<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Rich Fallat on Medium]]></title>
        <description><![CDATA[Stories by Rich Fallat on Medium]]></description>
        <link>https://medium.com/@richfallatjrshop?source=rss-e21153971cf3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*6pDkHN7oePR7mNRW</url>
            <title>Stories by Rich Fallat on Medium</title>
            <link>https://medium.com/@richfallatjrshop?source=rss-e21153971cf3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 19:09:22 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@richfallatjrshop/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Gravitas Crunch — Your reading, now on air]]></title>
            <link>https://medium.com/@richfallatjrshop/gravitas-crunch-your-reading-now-on-air-6b0452be6ae2?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/6b0452be6ae2</guid>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Wed, 16 Sep 2026 17:29:46 GMT</pubDate>
            <atom:updated>2026-10-10T10:49:13.700Z</atom:updated>
            <content:encoded><![CDATA[<h3>Gravitas Crunch: Your Sources Become the Show</h3><h3>Inside a source-first reader, research assistant, personal broadcast network, and on-device presentation studio.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bqUJgNW2xgxrUx6fIZttfA.png" /></figure><p>Hey, I’m Jett. That reading list? Let’s make it a show. Crunch brings articles, PDFs, Kindle books, and your favorite feeds to life with animated presenters. Watch or listen, with AI running on your device. Try it free. Unlock once. No subscription.</p><p>The idea behind Gravitas Crunch is simple: the things you are interested in should not be trapped in the format you found them in.</p><p>An article can be something you read. It can also become a briefing you listen to, a subject you investigate, a study guide you return to, or an animated presentation you share with someone else.</p><p>Those should not have to be separate projects.</p><p>Imagine opening a research paper, asking for the technical explanation, following a citation back to the relevant passage, and then asking for a version that makes sense to someone encountering the subject for the first time. When the explanation is right, you press Broadcast. A presenter delivers it on screen, accompanied by imagery from the source. You can replay the segment, save the video, or share it.</p><p>Or imagine doing none of that work deliberately. You open Robot Radio, choose a station, and let the publications you follow become a lineup of animated reports. Something catches your attention. You open the article and go deeper.</p><p>Both experiences belong to the same app.</p><p><strong>Gravitas Crunch is a source-first reading and research workspace that turns articles, feeds, long-form text, and your own explanations into things you can read, hear, watch, and keep.</strong></p><p>The reader is its foundation. Research gives you a way to work with the material. Robot Radio brings your feeds on air. Audiobook narration lets you hear the longer version. MindsEye gives the voice a visible performance. Broadcast turns that performance into a video.</p><p>The connective tissue is the source. You do not have to abandon it to change how you experience it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2jbB-NLpS0gyVyOjnOzqIA.png" /></figure><h3>“Don’t read. Watch.” does not mean abandoning reading</h3><p>“Don’t read. Watch.” is the shortest expression of Crunch’s broadcast idea. But there is a full reader underneath that sentence, and that is intentional.</p><p>Sometimes the original paragraph is exactly what you need. Sometimes you need a quick orientation before deciding whether to spend an hour with a paper. Sometimes you want the full article, but would rather listen. Sometimes you understand the material and want to explain it to somebody else.</p><p>I do not think an app should decide that only one of those relationships with information is legitimate.</p><p>Crunch is built around moving between them. You can begin with a brief and end up reading the source closely. You can begin with a long document and make a short presentation. You can keep an explanation as text rather than video. Watching is an option, not a requirement.</p><p>The important change is that research does not have to end at an answer sitting in a chat window. It can become something you deliberately shape and produce.</p><p><strong>Watch your sources. Shape the explanation. Keep the broadcast.</strong></p><h3>Start with something real</h3><p>Crunch starts with material you bring into it, follow, or discover — not a demand that you invent the perfect prompt before anything useful can happen.</p><p>You can paste a link, use Read from Clipboard, import a file, open a feed item, or explore through Discover Assistant. Supported reading inputs include articles, PDFs, EPUBs, plain text, and Markdown. The app also handles research-oriented sources and supported video and podcast pages, with available text and transcripts depending on the source.</p><p>That range matters because an interest rarely arrives in one neat format. An explanation might begin in a newsletter, point to a paper, lead to a reference article, and eventually become something you want to keep in your own notes.</p><p>Crunch includes source-specific handling for services and formats such as Medium, Substack, WordPress, Ghost, Wikipedia, Reddit, arXiv, PubMed, and DOI-linked research, alongside more general article extraction. YouTube, Vimeo, and podcast sources have their own handling rather than being treated as ordinary text pages.</p><p>The goal is not to make all these sources look as though they originated in the same place. It is to bring their usable content into a consistent workspace while preserving the route back to where they came from.</p><p>Discover Assistant is the outward-facing part of that workflow: finding more material to explore. Research Assistant is the inward-facing part: working with the source you have opened. Discovery helps answer “What should I look at?” Research helps answer “What is this saying, and what do I want to understand about it?”</p><p>You can also follow publications through RSS and Atom. That gives Crunch an ongoing supply of material rather than making every session begin with a new import. The same feeds can support ordinary reading and your personal Robot Radio lineup.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/941/1*bhzOB1dxOsUVHykwIqlU6A.png" /></figure><h3>The reader is infrastructure, not a waiting room</h3><p>A presenter is easy to notice. A good extraction layer is easier to overlook. But much of the app depends on what happens before a summary or voice is generated.</p><p>A URL is not yet a usable source. The app has to identify the kind of material it has received, retrieve what is available, and build a reading representation. Different source types need different treatment. Crunch’s extraction path uses specialized resolvers and a general web-extraction fallback rather than assuming one approach will handle every page.</p><p>That work establishes the material the rest of the app can use: the readable content, its structure, identifying information, and available media. Research, narration, and presentation all depend on that foundation.</p><p>I think of this as keeping the source usable throughout the experience. The reader is not merely a place where an article waits until AI replaces it with something shorter. It remains the place you return to when a detail matters.</p><p>Source access also has practical boundaries. A page may expose only part of an article. A feed may contain an excerpt. A publisher may not provide a transcript. Crunch works with the material available through the supported source; the word “supported” does not mean every page exposes identical content.</p><p>The same distinction applies to media. Playing a publisher’s podcast or displaying a supported video page is different from generating a Crunch broadcast. Podcast playback uses publisher-provided feed enclosures. Crunch’s generated broadcasts use text, locally produced narration, presenter assets, and available imagery; they are not downloads or conversions of the original YouTube, Vimeo, or Reddit video.</p><p>That separation keeps the product understandable: the original source is one thing, and the explanation or presentation you make from it is another.</p><h3>Crunching is about choosing a useful form</h3><p>A request to “summarize this” leaves a lot unspecified. A useful summary for someone making a quick decision is not necessarily a useful summary for someone learning the underlying method.</p><p>Crunch offers three principal research styles: <strong>Executive, Technical, and Study Guide</strong>. They give the initial explanation a purpose before you begin refining it. The research workflow also supports follow-up questions, outlines, source-linked citations, and contextual notes.</p><p><strong>Executive</strong> is for orientation. What matters most? What is the central argument? Which takeaways are worth carrying into the next conversation? It is a starting point when you want the high-level picture before committing to the detail.</p><p><strong>Technical</strong> moves toward methods, assumptions, and the explanation underneath the headline. It is the mode to choose when “what happened” is less useful than “how does this work?”</p><p><strong>Study Guide</strong> organizes the material around learning: concepts, definitions, relationships, and a structure you can revisit. The intent is to make a source useful for review rather than merely compress its word count.</p><p>These are not three different libraries or three disconnected assistants. They are different ways of approaching the same material.</p><p>For example, you might use Executive to decide whether a paper deserves closer attention, Technical to investigate the method, and Study Guide to organize the ideas you want to remember. You can also begin directly with the mode that matches your purpose.</p><p>The first result is a starting point. Conversation is how you continue directing the work.</p><h3>Long reads need a system, not just a bigger text box</h3><p>Long-form material is central to Crunch. That includes the practical engineering problem of working with documents that are larger than a single local model request can comfortably accommodate.</p><p>A language model has a limited context window: the amount of material it can consider in a session. Apple’s guidance for its on-device model includes splitting larger tasks into smaller model sessions and assembling their results. The model is one part of the solution; managing the document is another.</p><p>Crunch’s long-form pipeline divides the normalized source into bounded chunks, associates those chunks with their positions in the document, processes them with controlled concurrency, and restores their source order before building the final explanation. Completion order is not allowed to become document order.</p><p>The distinction is important. A later section might finish processing before an earlier one, but it still belongs later in the source. Likewise, the boundaries used to fit text through the model are processing boundaries, not a claim that the article naturally divides into those exact pieces.</p><p>This lets the app approach long material as a structured job rather than simply paste a large document into a prompt and hope for the best.</p><p>The architecture also separates this research work from the later task of preparing speech. Dividing a paper for analysis and dividing narration for playback are different operations with different purposes. One manages source context; the other manages delivery.</p><p>That separation is part of what allows reading, research, radio, and presentation to share a foundation without becoming the same feature.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B6E25PpV3Gl96_GHFDvVPg.png" /></figure><h3>Focus Mode: take a question through the source</h3><p>Sometimes you do not want a general overview at all. You already know what you are looking for.</p><p>Focus Mode applies a question or instruction across the source. In the long-form pipeline, that focus accompanies the individual chunk requests as well as the final explanation. It is not merely a new label placed on a generic summary after the work is done.</p><p>You might ask:</p><blockquote><em>What assumptions does this method depend on, and where does the author discuss its limitations?</em></blockquote><p>Or:</p><blockquote><em>Explain the parts that would matter to someone implementing this approach.</em></blockquote><p>Or:</p><blockquote><em>Identify the concepts I would need to understand before teaching this material to a beginner.</em></blockquote><p>Those are different reading intentions. Focus gives the app a way to carry an intention through the document rather than treating every sentence as equally relevant to your immediate task.</p><p>Afterward, you can ask follow-up questions, investigate a point, request a different explanation, or turn the useful result into an output. The work remains connected to the source that prompted it.</p><h3>Conversation is the editing interface</h3><p>A good explanation is not always the shortest one. It is the one suited to the person who needs it.</p><p>Crunch lets you change the tone, length, audience, and focus through conversation. You can request a concise briefing, a deeper technical treatment, a study segment, a more conversational explanation, or a product introduction.</p><p>That is where research begins to become authorship.</p><p>“Make it shorter” is one instruction. “Keep the technical explanation, but explain the terminology before using it” is another. “Rewrite this for a colleague who understands the problem but has not read the paper” produces a different kind of direction again.</p><p>The important thing is that you can articulate the editorial decision in ordinary language. You do not need to rebuild the context in a separate writing tool before you can work on how the explanation should sound.</p><p>The source remains close. Research output includes supported citations and provenance links that let you return to the relevant passages. Notes can stay attached to the active research conversation, and the results can be exported instead of remaining trapped in the session.</p><p>For me, that combination matters more than an answer simply sounding polished. I want the experience to encourage movement between the explanation and the evidence behind it.</p><p>You can read closely, ask questions, refine a response, and then decide what deserves to be presented. Broadcast comes after that editorial choice, not instead of it.</p><h3>Robot Radio: your interests, on air</h3><p>Robot Radio takes the feed library and turns it into a different kind of experience.</p><p>Instead of asking you to open every headline individually, it prepares source-based, anchor-style reports and plays them as a station-like lineup. These are locally narrated presentations generated from source text — not rebroadcast recordings supplied by the publications.</p><p>There are three main ways into that lineup.</p><p><strong>Now</strong> draws on fresh, eligible stories from the available feed library. It is the place to begin when you want to hear what is current in the material the app can access.</p><p><strong>My Station</strong> builds around the feeds you choose to follow. The personalization begins with your source selection: the publications and interests you decide belong in the mix.</p><p><strong>Curated stations</strong> provide another entry point, organized around areas such as news, technology, science, space, business, gaming, entertainment, and sports. You can channel surf rather than first construct an entire feed collection yourself.</p><p>I wanted the experience to feel like programming, not a queue of unrelated text-to-speech files. The selection, presentation, music, and ability to move between stories all contribute to that feeling.</p><p>MindsEye adds the visible broadcast: a presenter, studio environment, source imagery, and graphics. The report becomes something you can watch, while the underlying source remains available through Open Article.</p><p>That makes Robot Radio useful both as an experience in itself and as a way into deeper reading. A short report can tell you enough to move on. It can also reveal something you want to investigate.</p><p>The radio is not a dead end. It is another entrance to the reader and research tools.</p><h3>Music and voices are part of the presentation</h3><p>A broadcast has a tone as well as a transcript.</p><p>Robot Radio uses context-selected music from the app’s authored track library. Apple’s on-device model helps select a track appropriate to the story’s subject and tone. It selects music; it does not generate a new song for every article or borrow the soundtrack from an embedded publisher video.</p><p>I treat that distinction as part of the design. The music is an authored resource. The choice of where it belongs is contextual. A voice supplies the spoken report, while the surrounding presentation gives it pacing and atmosphere.</p><p>Voice selection is part of the listening experience too. Crunch includes voice controls and a Multi-Voice system that can vary the voice across content using stable assignments. In Robot Radio, the implementation associates voice selection with the story’s identity, rather than randomly changing the speaker every time another piece of audio becomes ready.</p><p>The purpose is a more coherent listening experience, not a claim that a synthetic presenter is the original author or a reporter who witnessed the events.</p><p>The article supplies the material. Crunch supplies a generated presentation of it. Keeping those roles distinct is compatible with making the presentation enjoyable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DgEQ3bPdtFBhIX6lYHvLVg.png" /></figure><h3>Audiobook mode: keep the long version</h3><p>A short radio report is not always what you want. Sometimes you want the article itself, delivered as speech.</p><p>Audiobook mode turns readable material into sectioned narration. The sections provide a navigable structure rather than one unbroken recording: you can move between sections, choose a voice, and retain the article’s identifying context while listening.</p><p>That gives Crunch three distinct relationships with spoken material.</p><p>Robot Radio prepares a shorter, broadcast-style report from a source. Audiobook narration lets you listen to the fuller text. Research Broadcast performs the explanation or transcript you deliberately select.</p><p>None makes the others redundant.</p><p>A radio report can introduce a subject. You can open the article and listen to the full treatment. Later, you might ask a question and turn the answer into a short video for someone else.</p><p>This is also separate from podcast playback. A podcast episode is publisher-provided media; an article audiobook is narration generated from reading material. Available podcast transcripts offer another route between listening and reading, but not every episode provides one.</p><p>The common idea is continuity. Switching to audio should not mean losing track of what you were reading or why it interested you.</p><h3>MindsEye began with a missing part of a conversation</h3><p>The original desire behind MindsEye was not “I should add an avatar.”</p><p>It was that I wanted to see the character on the other end of Turing, the conversation system in Gravitas Plague. I wanted to see the character working and delivering the response. The voice existed, but the experience felt incomplete without the visible performance.</p><p>That is a different problem from finding something decorative to place beside an answer.</p><p>A character sitting next to text can remain separate from the interaction. A character delivering the dialogue becomes part of how the interaction reaches you. The voice, mouth, expression, and pauses belong to one performance.</p><p>In Crunch, that insight changes the output. A research response is no longer only something you read or hear. It can become a segment with a presenter.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tBjHrpTTnx_oBVZC4ScwYw.png" /></figure><p><strong>MindsEye is a real-time procedural 2D presentation system.</strong> It combines authored character artwork, lip sync, subtle motion, layered environments, available source imagery, and broadcast graphics. The result is composed on the device rather than requested as a finished cloud-generated video.</p><p>The authored part matters. The character’s design, silhouette, wardrobe, and presentation establish a coherent visual identity. Runtime systems then bring that design into motion and connect it to the spoken material.</p><p>This is not asking a video model to reinvent the presenter in every shot. It is using a designed character as the basis for a continuing performance.</p><p>The visual system also supports different presenter styles, from realistic to more stylized treatments. Those are presentation choices, not different research engines. The voice and source identity remain tied to the active content while the visual package gives it a different aesthetic. I want room for personality without making the character selection more important than what the character is saying.</p><p>That also makes the system an interesting meeting point between art direction and software. The artwork determines what the character can convincingly express. The runtime determines when and how those expressions appear. Neither side can substitute for the other.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kUyVXvrHpsyjmkdB-8ahnA.png" /></figure><h3>The mouth should belong to the voice</h3><p>There is an important technical distinction inside MindsEye: a mouth opening whenever audio becomes louder is not the same thing as lip sync.</p><p>Crunch’s presentation system analyzes generated speech in relation to its spoken text and prepares timing information for the mouth performance. The playback system uses the position in the actual audio to select the corresponding visual state. The speech and the face are tied to the same performance, not driven by unrelated timers.</p><p>That relationship matters when playback pauses, resumes, or the display does not deliver every expected frame. The useful question is “Where are we in the spoken performance now?” rather than “How many animation updates have happened?”</p><p>The character is also only one part of the image. The full broadcast includes the studio, article imagery, framing, and graphics around the presenter. Crunch composes those layers into a scene rather than treating a transparent animated face as the entire product.</p><p>I care about that whole composition. A face may attract attention, but the image still needs hierarchy. The source imagery should have a place. The presenter should feel integrated with the set. The graphics should support the material rather than fight it.</p><p>This is where MindsEye becomes more than a demonstration of mouth movement. It is a presentation language for the app.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FQzJWvx9jI4yBfAfAqlEcQ.png" /></figure><h3>Broadcast: make something you can keep</h3><p>Robot Radio assembles a continuing lineup. Broadcast is a deliberate act of creation.</p><p>You choose a completed research response or a transcript, press Broadcast, and let MindsEye perform the selected material. The central flow is <strong>choose the text, watch the presentation, then replay, save, or share the video</strong>.</p><p>Selecting one response is useful when you have refined a particular explanation. Selecting the transcript is useful when the progression of questions and answers is itself the material you want to present.</p><p>The editorial work happens before the performance. You can shorten the explanation, change its audience, adjust the tone, or focus it around a specific question. Then you choose what to broadcast.</p><p>This is a meaningful boundary. Asking for a different explanation and producing a video of the chosen explanation are different operations. Pressing Broadcast should not be an invitation to invent a different argument behind your back.</p><p>The broadcast preparation path retains the selected content in ordered blocks, prepares its narration, and checks that the assembled narration covers that selected text. Its job is delivery, not a second unsolicited summary.</p><p>The result might be a study explanation, a technical briefing, an introduction to a product, or a short presentation of an idea. The format is useful because the explanation becomes portable. You can keep it for later or send someone the presentation, not just tell them which conversation to reconstruct.</p><h3>You can write the script yourself</h3><p>Crunch does not require every useful sentence to originate in an AI response.</p><p>The Notepad gives you an editable space associated with the source. You can write, revise, and save your own text, then copy it, share it as plain text or Markdown, export a PDF, or send it into Broadcast.</p><p>This matters because conversation and direct editing serve different purposes.</p><p>Sometimes you want to describe a change and have the assistant produce another version. Sometimes you know the exact sentence you want and would rather type it. Sometimes you already have a script and want the presentation system to perform it.</p><p>An explanation can begin with research and end with your own editorial pass. A commercial can begin as a script. A study segment can combine the material you have investigated with the phrasing that helps you remember it.</p><p>The app should support those choices rather than force everything through the same prompt-driven ritual.</p><p>That is also why I see Broadcast as a creation tool, not merely a more elaborate read-aloud button. You decide what the final material should say. The system gives it a voice and a visible performance.</p><h3>A finished video is not a recording of the app</h3><p>There is a difference between watching a presentation being prepared and watching the finished presentation.</p><p>A screen recording captures whatever happened while the app was running, including preparation waits and interface controls. The Broadcast design instead builds the saved video around the media program itself: the selected narration, intended pauses, character performance, scene elements, and presentation transitions. Generation waits are excluded from that finished timeline.</p><p>The distinction is easy to miss until you think about sharing. Someone receiving your explanation does not need to watch the app prepare it. They need the explanation.</p><p>Natural pauses belong to the performance. Time spent preparing the next portion does not.</p><p>The live presentation and the exported video are therefore related views of the same prepared material, rather than two unrelated attempts to generate the same script. Replay should bring back the presentation you made, not ask for a fresh interpretation.</p><p>This is one of the choices that turns a compelling demo into a useful output. The presentation can leave the workspace and still stand on its own.</p><h3>Your work should survive the session</h3><p>A reading and research app needs more than a successful first interaction. It needs a way to return.</p><p>Crunch includes saved sources for offline reading, History for revisiting material, and notes connected to research. Research conversations can be copied or shared as plain text, exported as Markdown, or made into a PDF. Broadcast adds video to those outputs rather than replacing them.</p><p>The appropriate format depends on the next task. Markdown may be useful in a writing workflow. A PDF may be the right way to preserve a structured explanation. A video may be more useful when you want someone to watch a segment. A note may be all you need.</p><p>The point is that the work is not disposable just because part of it happened through conversation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9IrqQM7SJqmusIkZvY8cRA.png" /></figure><h3>One workspace, different devices</h3><p>Crunch brings these workflows to iPhone, iPad, Mac, and Apple Vision Pro. The design is intended to suit those different contexts rather than insist that every screen show the same arrangement.</p><p>On a wider workspace, the organizing idea is straightforward: discover on the left, read or watch in the center, and research on the right. It places the source and the work you are doing with it close together.</p><p>On iPhone, the experience needs to prioritize the active task. The reading, listening, and broadcast surfaces have to remain useful without reproducing a desktop’s three-column density.</p><p>Liquid Glass is part of the visual language, but the interface still has a practical job: make the relationship between the source, the player, and the research understandable. The point is not transparency for its own sake. It is a workspace that can hold several modes without making them feel like unrelated applications.</p><p>The continuity here is the product model. It should not be confused with a promise that every local document or session automatically synchronizes between devices.</p><h3>On-device is an architectural choice</h3><p>Crunch uses Apple’s on-device Foundation Models capabilities for supported research, summaries, focused questions, and conversational rewrites. Speech is generated locally, and MindsEye’s animated presentation is composed on the device. These are distinct stages of the system, not one call to a service that sends back a finished show.</p><p>The sequence is roughly: acquire usable source material, prepare the explanation, prepare the spoken delivery, generate speech, align the visible performance, and compose the presentation. Broadcast adds the work of preserving that program as a shareable video.</p><p>Keeping those responsibilities separate matters. The reading layer should not depend on the presenter’s design. Changing a voice should not change the source. A new visual treatment should not require rebuilding the research workflow.</p><p>Local processing also makes resource management a real part of the product. Source preparation, language-model work, speech generation, audio playback, and animation all have to coexist on the same device. Preparing material ahead of playback and keeping completed pieces in order are therefore central engineering concerns, not optional polish. Crunch’s research and presentation architecture explicitly separates those jobs and bounds the work they can perform concurrently.</p><p>I do not want the interface to make the user think about that machinery. But it is part of what makes the experience possible.</p><h3>What privacy and availability mean in practice</h3><p>Core Crunch workflows do not require a Gravitas account. Your local library and saved research are there to help you keep working, and sharing is a deliberate action you take through the app’s export and system-sharing controls. Optional source-service connections are separate from requiring an account simply to use the reader.</p><p>“On-device” does not mean that retrieving the web requires no internet. Discovering material, refreshing online sources, loading publisher pages, and playing publisher-hosted media still involve those online sources. The important distinction is between acquiring the material and performing the supported local analysis, narration, and animation.</p><p>AI-assisted features require compatible hardware and Apple Intelligence to be enabled and available. Apple’s system model may need time to become ready, and availability depends on supported devices and regions. Reading and other non-assisted source workflows are not the same requirement as language-model-powered research.</p><p>Similarly, saving an article for offline reading is not a promise that every externally hosted image, video, podcast, or linked page has also been downloaded.</p><p>Those distinctions define the product more accurately than an absolute claim that everything is always offline. The aim is a local working environment with clear, purposeful connections to the material outside it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CeVxcTChAsth_QriniDsFw.png" /></figure><h3>What the whole experience makes possible</h3><p>The most useful way to understand Crunch is to follow an interest through it.</p><p>You might begin in Robot Radio, hear a report about a new graphics technique, and open the source. Technical mode helps you investigate the explanation. Focus carries a question about implementation tradeoffs through the material. You follow a citation back to the relevant passage, write a few notes, and make a short briefing for a colleague.</p><p>Or you might import an EPUB, work through its ideas in Study Guide mode, and refine one explanation until it matches the way you learn. You keep a text version for review and a narrated presentation for another time.</p><p>Or you might bring in your own product material, develop a script, edit the final wording in Notepad, and create a presenter-led introduction. The research tools help when you need them; direct writing remains available when you already know what you want to say.</p><p>These are examples of paths through the app, not separate products hidden behind separate imports.</p><p>You can stop wherever the work is useful. You do not have to broadcast every article. You do not have to summarize everything you listen to. You do not have to turn an ordinary reading session into a production exercise.</p><p>The advantage is having those possibilities connected when you need them.</p><h3>The idea that connects everything</h3><p>I am interested in the point where information becomes an experience — and where that experience gives you something useful to do next.</p><p>The reader gives you the material. Research gives you ways to question it. Citations give you a route back. Radio gives your interests a running program. Audiobook narration gives long reads another form. MindsEye gives the voice a performance. Broadcast gives that performance a life outside the app.</p><p>None of those pieces explains Crunch completely on its own.</p><p>A reader without the rest would leave the material in roughly the form you found it. A presenter without the reader and research would have much less to say. A conversation without an output would leave the work waiting in a transcript.</p><p>The project is the connection between them.</p><p>I want you to be able to encounter something interesting, understand it more deeply, decide how it should be explained, and make a version worth keeping.</p><p>Not just another saved link. Not just another answer.</p><p><strong>Your sources. Your questions. Your explanation. Your broadcast.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LUXFR-OQZ8xByh-1xZqp5A.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6b0452be6ae2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Don’t Read. Watch. I Built a Personal Animated Broadcast Network.]]></title>
            <link>https://medium.com/@richfallatjrshop/dont-read-watch-i-built-a-personal-animated-broadcast-network-1425b3aa7097?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/1425b3aa7097</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[news]]></category>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Wed, 09 Sep 2026 04:34:08 GMT</pubDate>
            <atom:updated>2026-09-09T04:34:08.992Z</atom:updated>
            <content:encoded><![CDATA[<h3><em>Gravitas Crunch turns the sources you follow — and the research you create — into real-time animated broadcasts and video briefings you can keep.</em></h3><p>I left Gravitas Crunch running on my Mac for about an hour. Presenters delivered stories, article images cycled beside them, and the broadcast kept going. The whole thing was purring like a kitten.</p><p>What made that satisfying was not just that the software worked. It was that I wanted to keep watching it.</p><p>When I first wrote about Crunch, I described a personal broadcast for the real web. It began as a reader, grew into a research workspace, and eventually became Robot Radio: a station built from real publications and the sources I wanted to follow.</p><p>The latest update makes that idea visible.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FCgCQHTQSrEc%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DCgCQHTQSrEc&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FCgCQHTQSrEc%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/96a7b05e58edefd142783df7aecdefeb/href">https://medium.com/media/96a7b05e58edefd142783df7aecdefeb/href</a></iframe><p>I added <strong>Mind’s Eye</strong>, a real-time procedural 2D animation system that turns the broadcast into a performance. Then I connected that performance to research, so a summary or a response can become a finished video segment of its own.</p><p>The shortest description of the product is now the simplest:</p><p><strong>Don’t read. Watch.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3B97BP3nsYUjpmixzAGQGg.jpeg" /><figcaption><em>The broadcast, the original source, and the research tools belong to the same workspace.</em></figcaption></figure><h3>The reader finally became a stage</h3><p>The original reader already had the beginnings of this. Article imagery could rise into a lightbox above the text. There was movement, glass, glow, and an effort to make spending time with a source feel more engaging than looking at a static document.</p><p>Mind’s Eye gives that stage a performer.</p><p>An animated presenter occupies an authored broadcast set. The source’s images appear beside the character. A lower third identifies the story and publication. Locally generated speech drives the delivery, while blinking, mouth animation, subtle movement, and layered parallax keep the character alive.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kUyVXvrHpsyjmkdB-8ahnA.png" /></figure><p>The result is not just narration with a photograph attached. It is a composed shot.</p><p>That distinction matters to me as an artist. Where the presenter stands, how much space the article image receives, how the background supports the foreground, and how the graphics sit inside the frame all affect whether the experience holds together. A book cover, a phone screenshot, and a landscape photograph have very different shapes. The presentation needs to accommodate all of them without looking as though a new piece of software has been assembled for each story.</p><p>I want the visual system to be recognizable even when everything being discussed changes.</p><h3>It is animation assembled in real time</h3><p>Mind’s Eye uses authored artwork and procedural performance. The character, expressions, set, and graphic treatment establish the visual identity. The runtime combines those elements with the speech and source imagery to produce the moving presentation.</p><p>The artwork is prepared. The particular broadcast is not a prerecorded clip.</p><p>That is the part I find exciting. I can bring new material into the system without authoring a new animation from scratch or sending each segment to a remote video-generation service. The same visual language can support an article, a radio report, a longer narration, or a research response.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2jbB-NLpS0gyVyOjnOzqIA.png" /></figure><p>Coming from character look work, I do not regard the controlled nature of the system as a compromise. It is a way to put the effort where the audience will see it. I care about the silhouette, the expression, the relationship between layers, and whether the character belongs in the shot. I do not need to solve every possible kind of animation to make a convincing presenter.</p><p>The goal is a sustained performance with a deliberate look. The system gives my art direction somewhere to live beyond a single finished image.</p><h3>The most surprising feature came from research</h3><p>I expected to enjoy the animated station. What surprised me was how much I liked broadcasting the things I was making inside Research Assistant.</p><p>Crunch already lets me work with long-form material in different ways. Executive mode gives me the main ideas. Technical mode concentrates on methods, assumptions, and detail. Study Guide mode organizes concepts for review. Focus lets me bring a particular question to the source rather than compressing everything with the same priorities.</p><p>Previously, the obvious outputs were text, spoken playback, and exported notes. Now a response can become a broadcast segment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CeVxcTChAsth_QriniDsFw.png" /></figure><p>I can choose one answer and press <strong>Broadcast</strong>. The selected text becomes the narration, the presenter performs it, and the article’s available images cycle through the composition. I can also work with an entire transcript instead of a single answer.</p><p>That changes the role of the research workspace. I am no longer only asking the system to help me understand something. I can use that understanding to make something I can watch, revisit, or share.</p><p>A research artifact does not have to look like a document.</p><h3>The conversation becomes the script editor</h3><p>The next step feels almost obvious once you try it: ask for a different version, then watch that version.</p><p>Imagine starting with a product landing page and asking:</p><blockquote><em>Make this a short introduction for someone who has never heard of the product. Lead with what they can do with it.</em></blockquote><p>Then, after hearing the delivery:</p><blockquote><em>Cut the explanation in half. Keep the strongest benefit and make the ending more direct.</em></blockquote><p>Or take the same material in another direction:</p><blockquote><em>Turn this into a concise technical briefing. Explain the mechanism instead of selling the benefit.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P8HuS-jBvl3LmzpDa2B_Mw.png" /></figure><p>The source stays the same. The editorial instruction changes. The broadcast gives me a way to judge the result.</p><p>This is especially interesting for commercials and promotional copy. A sentence can look good on the page and feel cumbersome when someone performs it. Another line can seem almost too simple until the voice and presentation give it impact. I can hear what needs work, return to the conversation, and revise that part.</p><p>Each response becomes a possible take. I do not have to use the whole conversation or replace the version I liked. I can choose the particular response that deserves to become the segment.</p><p>There is still an important distinction between asking for a “30-second version” and getting exactly thirty seconds of speech. The requested length is an editorial target; the generated audio determines the actual runtime. But the interaction itself is powerful: shorter, clearer, more technical, more direct, less promotional. These are ordinary instructions, not video-editing operations.</p><p>The conversation edits the script. Mind’s Eye supplies the performance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dru03Rcba8YJEm30FY3SzA.png" /></figure><h3>Keep the performance, not the waiting</h3><p>The broadcast research artifact also needs to behave like a finished piece of media.</p><p>When the segment ends, there is a familiar end state: replay it, share it, or save the video through the system’s export flow. The result can leave the application instead of existing only as a live session.</p><p>The important detail is what the video does not contain.</p><p>A device may need time to calculate the next piece of speech. Someone may pause playback. Those delays belong to operating the software, not to the presentation. The saved segment should contain the performance without the dead time spent producing it.</p><p>That does not mean flattening every pause in the voice. Breaths, punctuation, and the space between ideas are part of a good delivery. The distinction is between a presenter pausing for meaning and a computer pausing because it has not finished working.</p><p>The artifact follows the media timeline, with a smooth opening and ending, rather than preserving every second that passed during the session.</p><p>I think this makes video useful as a research output rather than just a novelty. A short briefing can be something you save for yourself. A source-based explanation can be something you send to a colleague. A study segment can be something you return to later. And a promotional piece can be something you refine through conversation before keeping the version that plays best.</p><p>The live broadcast is something you tune into. The artifact is something you made.</p><h3>The source still has to be there</h3><p>None of this would mean much to me if the attractive presentation separated the information from its source.</p><p>Crunch began with the unglamorous work of opening material properly. Articles, PDFs, EPUBs, feeds, and research papers arrive with different structures. A newsletter is not a journal article. A book is not a podcast page. Some sources supply clean text and useful images; others need more careful handling.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K-i435s06hVkd8Gju9y4Gw.png" /></figure><p>The resolver layer gives those inputs a usable home. It handles supported publishing platforms and formats, including sources such as Medium, WordPress, Substack, Wikipedia, arXiv, and PubMed. Availability still depends on what a source makes accessible. A broadcast interface does not make every page on the internet equally easy to extract.</p><p>Discovery belongs to that foundation too. Finding a feed is different from finding a paper or a book. I want the material entering the system to be chosen with intention, because a polished delivery cannot compensate for a poor source.</p><p>Once a source is inside Crunch, it remains available. Research answers can carry citations that take me back to the supporting passage. I can move from the explanation to the evidence instead of accepting the presenter as an authority simply because the shot looks convincing.</p><p>The model can still make mistakes. Citations give me a route to check the work; they are not a promise of infallibility.</p><p>This matters even more when the output looks like a broadcast. An AI-written summary is not the publisher’s own report, and the presence of the publisher’s imagery does not make the synthetic presenter their spokesperson. The source, the generated explanation, and my presentation system are different things. I want that relationship to remain understandable.</p><h3>Apple Intelligence helps organize the production</h3><p>In the iOS 27 version, I use Apple’s Foundation Models throughout the workflow: dividing long material into workable sections, writing summaries in different styles, segmenting speech utterances, and selecting music according to an article’s context.</p><p>Those jobs are related, but they are not interchangeable. Understanding a long source is different from preparing a sentence that will sound natural when spoken. Choosing a suitable musical transition is different again.</p><p>The music is selected from authored material; Foundation Models is not generating the soundtrack. The model works inside a system with rules, assets, and editorial choices I have designed.</p><p>The current speech implementation uses PocketTTS for local voice generation. Mind’s Eye supplies the visual performance. These parts cooperate, but I do not want a user to need a diagram of them before enjoying the result.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t8u6vR2UU0Yg1ZaWbLSjyg.png" /></figure><h3>Audiobook mode took the same foundation in a different direction</h3><p>Radio is discovery. Audiobook mode is depth.</p><p>Once a story catches your attention and you open the source, there are times when you do not want a radio segment. You want the full thing. You want the article, the essay, the paper, the EPUB, the long-form piece, and you want it narrated in a way that lets you stay with it.</p><p>That required a different orchestration pipeline.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yKMTsLl9eZGwloXsLN4uew.png" /></figure><p>Local generation matters to the product I want to build. It lets speech and supported research analysis happen on the device instead of making every response depend on a remote inference service. The app still needs network access to retrieve online sources, and the Apple Intelligence features depend on compatible hardware and settings. The distinction is that finding the source online and generating the presentation locally are separate jobs.</p><p>The interesting result is not a model name in a settings screen. It is the source becoming a coherent show.</p><h3>Radio for discovery, long-form playback for depth</h3><p>Robot Radio remains the easiest way to turn Crunch on and let it run.</p><p>A station can move through the subjects and publications I follow, packaging stories for listening and now presenting them visually. The experience can be broad, or it can stay close to a particular interest. The sources determine what the channel has to say.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DgEQ3bPdtFBhIX6lYHvLVg.png" /></figure><p>But a radio report and a full narration are not the same thing. Sometimes I want a quick understanding of a story. Sometimes I want to spend time with the original article or book without reducing it to a summary.</p><p>That is why the deeper reading and audiobook paths still matter. Mind’s Eye adds another presentation layer; it does not require everything to become a short news script.</p><p>The same source can support a quick report, a longer narration, a focused research conversation, and a finished briefing. I can choose the form that suits what I am doing.</p><h3>A workstation, not a collection of disconnected tools</h3><p>I spend a lot of time in complex creative software. Blender, Maya, and Houdini are not built around pretending that sophisticated work has only one step. They give different operations their own places while keeping them connected to the thing you are making.</p><p>That sensibility influenced Crunch.</p><p>On the Mac, discovery and navigation live on the left. The source and broadcast occupy the center. Research lives on the right. The performance can take most of my attention, but the material and the tools remain close.</p><p>I do not have to leave the reader for a chatbot, leave the chatbot for a speech application, and then assemble the result in a separate presentation tool. The systems are woven into one workspace.</p><p>That does not mean everybody needs to operate all of them at once. Sometimes I simply want to watch. At other times I want to investigate a claim, change the explanation, or create a segment. The interface should let me change tasks without losing the source I was working with.</p><p>The other form factors have their own needs. iPhone needs immediacy. iPad gives the material more room. Apple Vision Pro gives the visual presentation a different sense of presence. I want a consistent medium across them, not one desktop screenshot stretched onto every display.I designed each Apple platform as its own experience</p><p>Because the medium changes, the form factor has to change too.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S9vpn1VawgnvfJYcQSnlvQ.jpeg" /></figure><p>Gravitas Crunch is multiplatform, but I never wanted that to mean one interface stretched across every device. Each platform got its own care.</p><p>On <strong>iPhone</strong>, the app feels intimate and immediate. It is the quickest way to turn on the station, open a story, and keep moving.</p><p>On <strong>iPad</strong>, the reader has room to breathe. It becomes a more relaxed canvas for reading, browsing, and shifting between listening and source work.</p><p>On <strong>macOS</strong>, the product becomes a workstation. Discovery, source reading, and research can live together on screen in a way that makes the whole architecture visible at once.</p><h3>What I built changed while I was building it</h3><p>The first version of this story ended with a personal broadcast. That was the destination I could see at the time.</p><p>Mind’s Eye has expanded what that means. The broadcast is now animated, and the research can become part of the program rather than sitting beside it as a separate output.</p><p>I can watch something I found. I can ask a question about it. I can turn the answer into a different kind of explanation, hear how it plays, and keep the result as a video.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dfQ5sECfnvw4c3BkFfB1lA.jpeg" /></figure><p>That is why I no longer think of this as merely a reader with an assistant or a radio app with a character. I think of it as a personal animated broadcast network, with a research workspace and a way to make programs inside it.</p><p>The publications, documents, books, and questions are still the reason to be there. The animation gives them a new form. The artifacts give that form somewhere to go.</p><p>I started by trying to make information easier to live with. Now I can make a show from it.</p><p><strong>Gravitas Crunch. Don’t read. Watch.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1425b3aa7097" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gravitas Plague]]></title>
            <link>https://medium.com/@richfallatjrshop/gravitas-plague-72af6ed8172b?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/72af6ed8172b</guid>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Sun, 31 May 2026 17:35:30 GMT</pubDate>
            <atom:updated>2026-09-10T00:24:30.141Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8RvEW5cq3nOjZjuBvM6kKg.png" /></figure><h3>The emergence of ATNV-15</h3><p>The official name came late.</p><p>By the time the world had a name for it, the virus had already moved through caves, farms, shelters, hospitals, airports, and cities.</p><p>The first reports called it animal aggression.</p><p>Then they called it drug psychosis.</p><p>Then they called it a rabies variant.</p><p>Then they called it social collapse.</p><p>Only after the first hospital wing went silent did the name become official.</p><p><strong>Amygdalo-Thalamic Neurotropic Virus 15.</strong></p><p>ATNV-15.</p><p>Most people never used that name.</p><p>The public called it something simpler.</p><p><strong>The Gravitas Plague.</strong></p><p>Not because it was plague in the old sense. It was not the Black Death. It was not a bacterium. It did not move like the medieval stories. It did not announce itself with swollen glands and certain death.</p><p>It was quieter than that.</p><p>Stranger.</p><p>Crueler.</p><p>It did not always kill the body.</p><p>It changed what was left inside.</p><p>The first samples came from a cave system in the southern hemisphere. The site had no famous name. It was one of those places that existed on old maps as a line, then a gap, then a warning. Deep limestone chambers. Wet air. Mineral walls. Colonies of bats packed into the dark by the thousands.</p><p>The expedition was routine on paper.</p><p>A biosurveillance team had been sent to catalog unknown viruses in bat populations. It was the kind of work that sounded heroic in grant language and reckless in hindsight. The scientists were not looking for disaster. They were looking for possibility. Future vaccines. Future warnings. Future papers. The modern world had learned to fear what it did not sequence, so teams went deeper and deeper into places no one had any business disturbing.</p><p>In the cave, the bats looked healthy.</p><p>That was the first mistake.</p><p>The animals were alert. Responsive. Normal. They moved in enormous waves when disturbed, and then settled back into the ceiling like living smoke. There were no mass die-offs. No visible lesions. No obvious sign that anything inside the colony had changed.</p><p>But the sequencing data came back wrong.</p><p>Not impossible.</p><p>Wrong.</p><p>The viral fragments did not match the expected families cleanly. They carried familiar shapes, familiar tricks, familiar protein behavior, but arranged in ways the analysts did not like. It looked like a virus that had borrowed from several old playbooks, then learned a new target.</p><p>The fifteenth isolate was the one that mattered.</p><p>At first, it was just a line in a database.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-opsWh_XbmhjPFXflEHc-Q.png" /></figure><p>Sample 15.</p><p>Bat saliva. Southern cave system. Unknown neurotropic profile.</p><p>The sample was archived, flagged, and sent through the normal chain.</p><p>The first lab note described it as having an unusual affinity for mammalian neural tissue.</p><p>The second lab note used a phrase that should have stopped everything.</p><p><strong>Preferential activity near rapid thalamo-amygdala signaling pathways.</strong></p><p>That phrase did not mean much to the public.</p><p>It meant everything to the people who understood fear.</p><p>The thalamus is one of the brain’s great switching stations. It receives information from the senses and routes it forward. The amygdala is one of the brain’s ancient threat detectors. It watches for danger before reason has time to speak.</p><p>In a healthy person, those pathways help create the first instant of fear.</p><p>The flinch before thought.</p><p>The recoil before language.</p><p>The animal warning that says, do not touch that, do not go there, do not stand still.</p><p>ATNV-15 did not simply inflame the brain. It did not just cause fever dreams or confusion. It appeared to attack the fast fear circuit itself.</p><p>The early theory was simple and horrifying.</p><p>The virus did not create courage.</p><p>It removed the brake.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YbyMNn8pRfI92aL_0by3GQ.jpeg" /></figure><p>Fear did not disappear cleanly. It broke apart. The infected still reacted to sound, light, pain, motion, and hunger. But the normal hierarchy collapsed. A person could stare into an oncoming car without stepping away. A wounded animal could charge a fire. A patient could bite through their own tongue and keep moving.</p><p>The infected did not become fearless in any noble sense.</p><p>They became ungoverned.</p><p>The first mammal cases outside the cave were dismissed.</p><p>A ranch dog attacked its owner and ignored two broken legs.</p><p>A fruit bat colony abandoned its normal roost and entered a town in daylight.</p><p>Feral cats began fighting in alleys until they died of exhaustion.</p><p>Livestock stopped retreating from noise, pain, or flame.</p><p>The reports were scattered, local, and easy to explain away.</p><p>Heat.</p><p>Poison.</p><p>Parasites.</p><p>Drought.</p><p>Human stupidity.</p><p>Then came the eyes.</p><p>In every species, the eyes changed before the behavior became impossible to ignore.</p><p>Not always cataracts in the medical sense, but a cataract-like haze. A milky film. A dull whitening that made the animal look half-blind, even when it still tracked movement. Hunters described deer with ghost eyes. Shelter workers described dogs that stared through people. Nurses described patients who looked awake but unreachable.</p><p>The eyes became the first symbol of the Gravitas Plague.</p><p>Clouded.</p><p>Reflective.</p><p>Wrong.</p><p>Inside the infected brain, the decline seemed uneven.</p><p>Language degraded early in humans. Not always at once. Some patients could speak in fragments for days. Others repeated phrases. Others answered questions with old memories that had nothing to do with the room they were in.</p><p>One man kept asking for his mother while trying to claw through a glass door.</p><p>One woman calmly recited a grocery list as she attacked an orderly.</p><p>Another patient sang half of a childhood song until her voice gave out.</p><p>Doctors first assumed hallucination.</p><p>Then psychosis.</p><p>Then intoxication.</p><p>That was the second great mistake.</p><p>The first human clusters appeared among people society had already trained itself not to see.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rq11-d2B3TdFBTilbYGBeg.jpeg" /></figure><p>The unhoused.</p><p>The addicted.</p><p>The undocumented.</p><p>The isolated elderly.</p><p>The people already living at the edge of hospitals, underpasses, shelters, detention centers, and emergency rooms.</p><p>They were the first to be infected and the last to be believed.</p><p>When they became incoherent, no one was surprised.</p><p>When they became violent, no one asked why.</p><p>When their eyes turned cloudy, it was called exposure, infection, malnutrition, or drugs.</p><p>The pathogen hid inside existing cruelty.</p><p>For weeks, maybe months, the world mistook the first human outbreak for a behavioral crisis.</p><p>Police reports described excited delirium.</p><p>Hospitals described psychiatric overflow.</p><p>Shelters described attacks.</p><p>Ambulance crews described bites.</p><p>Jails described inmates who would not sleep, would not answer, and would not stop moving.</p><p>In every report, the same details repeated.</p><p>Loss of speech.</p><p>Loss of restraint.</p><p>Loss of fear.</p><p>Loss of ordinary pain response.</p><p>Clouded eyes.</p><p>Fixation on motion.</p><p>Unprovoked aggression.</p><p>Animal-like posture.</p><p>A strange, prolonged survival state.</p><p>That last part made the virus worse.</p><p>ATNV-15 did not kill reliably or quickly.</p><p>Some infected died within days from injury, exposure, dehydration, secondary infection, or violence. Others survived for weeks. Some lasted longer. Their bodies entered a critical altered state, damaged but persistent. Fever came and went. Hunger changed. Sleep became fragmented. Pain response weakened. Social behavior collapsed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6cVrTa5SzAPcj4gEHz_RyQ.png" /></figure><p>They were not dead.</p><p>That mattered.</p><p>They were not corpses.</p><p>They were living mammals with broken governance.</p><p>The virus did not resurrect.</p><p>It reduced.</p><p>It stripped away the upper layers first. Speech. Shame. Caution. Planning. Social recognition. Then it pulled behavior downward into older systems. Territorial defense. Hunger. Dominance. Pursuit. Startle. Contact. Rage without purpose.</p><p>People called them zombies because people need old words for new fear.</p><p>But the infected were not undead.</p><p>They were still alive in the most terrible way.</p><p>The cave isolate spread first through bats.</p><p>Then through the animals that lived near them.</p><p>Then through the animals that ate what they touched.</p><p>Then through bites, blood, saliva, mucus, and contaminated bodily fluids.</p><p>The virus did not need to be perfectly airborne in humans to become a global threat. It only needed the right chain. Bats moved it across regions. Mammals amplified it. Humans misunderstood it. Cities concentrated it.</p><p>By the time the first national alert was issued, ATNV-15 had already crossed borders.</p><p>The official advisory was cautious.</p><p>Possible neurotropic zoonotic pathogen.</p><p>Avoid contact with aggressive mammals.</p><p>Report unusual animal behavior.</p><p>Do not handle bats.</p><p>Seek medical care after bites.</p><p>The public barely noticed.</p><p>There were always warnings.</p><p>There were always diseases somewhere.</p><p>There were always videos of strange animals, strange people, strange violence.</p><p>The first footage to go viral showed a man standing in the middle of a highway at night. Cars swerved around him. He did not move. His eyes reflected white in the headlights. When a truck horn blasted inches from his body, he turned toward the sound and walked into it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ch6SZPvnGjctxpJncodBpg.png" /></figure><p>The second video showed a horse charging into a burning barn.</p><p>The third showed a woman in an emergency room pressing both hands against a glass partition, calmly mouthing words no one could hear while blood ran down her wrists.</p><p>The fourth video was removed.</p><p>Then copied.</p><p>Then removed again.</p><p>Then there were too many to remove.</p><p>The name <strong>Gravitas Plague</strong> began as a rumor.</p><p>No one agreed where it came from. Some said it came from a leaked research group name. Some said it came from the way the infected seemed pulled by a new gravity, drawn toward motion, sound, heat, and living bodies. Some said it came from the feeling in the room when one of them entered.</p><p>A weight.</p><p>A pressure.</p><p>A sudden heaviness in the air.</p><p>Whatever the origin, the name stuck because it felt true.</p><p>When an infected host was nearby, the room changed.</p><p>People spoke softer.</p><p>Animals hid.</p><p>Lights seemed colder.</p><p>The world felt heavier.</p><p>There was no cure.</p><p>There were treatments for symptoms. Sedatives, restraints, antivirals, anti-inflammatory protocols, experimental neural stabilizers. Some worked briefly. Most failed. A few patients returned to partial lucidity for minutes or hours, then disappeared again behind the haze.</p><p>The deeper problem was not simply infection.</p><p>It was damage.</p><p>Once the thalamo-amygdala pathway was compromised, once the fear response had been stripped and the behavioral systems destabilized, stopping viral replication did not necessarily restore the person. The brain was not a switch. It was a landscape after fire.</p><p>Some families refused to accept that.</p><p>They played old music.</p><p>They brought photographs.</p><p>They spoke through glass.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xMiZ9GRi53dgTlcwGUUYTw.png" /></figure><p>They said names again and again.</p><p>Sometimes the infected reacted.</p><p>A turn of the head.</p><p>A twitch near the mouth.</p><p>A hand against the window.</p><p>Enough to destroy the living.</p><p>Enough to make containment feel like murder.</p><p>That was the moral wound of the Gravitas Plague.</p><p>The infected were not gone enough.</p><p>And they were not present enough.</p><p>Governments tried to classify hosts by stage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RP1e-LDZ0GU72gB5o1h8uQ.jpeg" /></figure><p>Stage One was confusion. Light sensitivity. Anxiety without explanation. Poor sleep. Delayed speech. Subtle eye haze.</p><p>Stage Two was fear collapse. Risk blindness. Loss of social boundaries. Fixation. Irritability. Reduced pain response.</p><p>Stage Three was behavioral regression. Language fragmentation. Aggression. Pursuit behavior. Animal posture. Clouded eyes.</p><p>Stage Four was critical persistence. Severe neurological impairment. Unstable metabolism. Extreme threat response. Possible long survival.</p><p>The classifications helped hospitals.</p><p>They did not help the people in the room.</p><p>When someone you loved reached Stage Three, the chart meant nothing.</p><p>The first containment zones were built around hospitals, shelters, and animal control facilities. Then farms. Then airports. Then entire districts.</p><p>But the virus had already taught the world a cruel lesson.</p><p>It did not begin where people were watching.</p><p>It began where people had stopped looking.</p><p>In abandoned tunnels.</p><p>In overcrowded shelters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bdLHxF0nLzLJBPhYGlVrKQ.jpeg" /></figure><p>In rural clinics.</p><p>In cages.</p><p>In camps.</p><p>In homes where nobody checked.</p><p>In caves where science went looking for the future and found something older than fear.</p><p>By the second year, the phrase <strong>Spillover Zero</strong> referred to the original cave event. The first breach. The first known crossing. The first moment the world could have turned back and did not.</p><p>No one agreed on who was responsible.</p><p>Some blamed the scientists.</p><p>Some blamed governments.</p><p>Some blamed the companies that funded pathogen discovery.</p><p>Some blamed the people who ignored the early victims.</p><p>Some blamed the animals.</p><p>The truth was less useful.</p><p>ATNV-15 was not evil.</p><p>It had no plan.</p><p>It found a route through mammalian life and followed it.</p><p>The horror came from what it revealed.</p><p>Civilization was not as deep as people thought.</p><p>Fear was not weakness.</p><p>Fear was architecture.</p><p>Fear kept hands from flames.</p><p>Fear kept animals from cliffs.</p><p>Fear kept a person from biting, charging, chasing, striking, and screaming until the body failed.</p><p>When the virus damaged that architecture, what remained was not freedom.</p><p>It was collapse.</p><p>The infected became mirrors of the mammal underneath the human.</p><p>Hungry.</p><p>Territorial.</p><p>Reactive.</p><p>Lost.</p><p>The cataract haze became the mark everyone feared. In poor light, it was easy to miss. In direct light, it was unmistakable. The eyes looked filmed over, as if the person had been pulled inward and something older was looking out through fogged glass.</p><p>Containment manuals warned responders not to mistake stillness for calm.</p><p>Stage Three hosts could stand motionless for long periods. Some faced walls. Some crouched in corners. Some appeared to listen to sounds no one else heard. They often remained still until triggered by motion, voice, proximity, or direct attention.</p><p>That detail became important.</p><p>Attention could wake them.</p><p>Not in a mystical sense.</p><p>In a predatory sense.</p><p>A host that seemed dormant could turn when observed. A head could pivot toward the source of movement. A body could follow the angle of a voice. A hand could reach before the mind behind it understood why.</p><p>People began to say the infected knew when they were being watched.</p><p>That was not scientifically accurate.</p><p>But it was close enough to be terrifying.</p><p>The later containment simulations were built around that principle.</p><p>Do not assume an idle host is safe.</p><p>Do not assume silence means absence.</p><p>Do not assume distance means control.</p><p>If a host turns toward you, create space.</p><p>If a host closes distance, interrupt forward motion.</p><p>If bodily fluid exposure occurs, report immediately.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yNBZitEehjkNSFVg_GiodA.png" /></figure><p>If the eyes cloud, isolate.</p><p>If speech fails, do not negotiate.</p><p>If fear is gone, the person you are speaking to may no longer be the person making decisions.</p><p>The Gravitas Plague did not end the world in one explosion.</p><p>It did something worse.</p><p>It made every room uncertain.</p><p>Every cough, every animal sound, every figure at the edge of the road became a question.</p><p>Is that person sick?</p><p>Are they high?</p><p>Are they afraid?</p><p>Can they still stop themselves?</p><p>Do they know I am here?</p><p>Do they know what they are doing?</p><p>Do they know who they were?</p><p>And beneath all of those questions was the one nobody wanted to ask.</p><p>If fear can be removed, how much of being human depends on it?</p><p>The first outbreak began in a cave.</p><p>The second began in the places people ignored.</p><p>The third began everywhere else.</p><p>By then, the official posters were simple.</p><p>White background.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ho8U82OdBWAOd8P1HSK05A.jpeg" /></figure><p>Black text.</p><p>One cloudy eye.</p><p>One sentence.</p><p><strong>The body may survive. The person may not.</strong></p><p>That was the Gravitas Plague.</p><p>Not death.</p><p>Not madness.</p><p>Not rage alone.</p><p>A disease of the oldest warning system in the mammal brain.</p><p>A virus that did not make monsters from nothing.</p><p>It found the animal.</p><p>It removed the fear.</p><p>And then it let the body keep walking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p3-yBKbM6aEMj_Q61CFV2Q.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=72af6ed8172b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gravitas Kitchen: The AI Food Agent That Gives Your Kitchen Memory]]></title>
            <link>https://medium.com/@richfallatjrshop/gravitas-kitchen-the-ai-food-agent-that-gives-your-kitchen-memory-b58db5051884?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/b58db5051884</guid>
            <category><![CDATA[ios-development]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ios]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[grocery-shopping]]></category>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Tue, 12 May 2026 01:57:43 GMT</pubDate>
            <atom:updated>2026-09-09T03:43:33.103Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LoFLpDyvMgecdBRNcGhcnw.png" /></figure><p>Most food apps start with the recipe.</p><p>Gravitas Kitchen starts with the food you already have.</p><p>That sounds like a small difference, but it changes the entire product.</p><p>A normal recipe app asks, “What do you want to cook?”</p><p>A grocery app asks, “What do you want to buy?”</p><p>Gravitas Kitchen asks something more useful:</p><p><strong>What is already in your kitchen, what can you make with it, and what do you need to get the rest of the way there?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PazUdF_qdXXr4Xh2iuuwVw.png" /></figure><p>That is the core idea.</p><p>Gravitas Kitchen is an AI food agent that helps keep inventory of the food you have, provides recipe discovery, keeps a cookbook, and procures items in your Shopping List through delivery and curbside pickup.</p><p>It is not trying to be a smart fridge. It is not trying to be a nutrition app. It is not trying to be another recipe website with prettier cards.</p><p>It is trying to solve a very normal, very human problem:</p><p>You have food.<br>You still do not know what to cook.<br>You buy duplicates.<br>You forget what is in the fridge.<br>You order takeout anyway.</p><p>Gravitas Kitchen is built around that loop.</p><p>Inventory becomes memory. Recipes become decisions. The Shopping List becomes action.</p><p>And AI is what makes the whole thing feel less like data entry and more like a useful agent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L05A5GcxM9RxycCQSYKtfQ.png" /></figure><h3>The kitchen is a memory problem</h3><p>Most people do not need more recipes.</p><p>They need context.</p><p>A recipe for chicken broccoli rice bowls is not useful in isolation. It only becomes useful when you know whether you have chicken, broccoli, rice, soy sauce, garlic, and green onions.</p><p>A shopping list is not useful in isolation either. It only becomes useful when it knows whether “soy sauce” is something you buy often, whether you already have it, and whether it should be matched to a specific product at a specific store.</p><p>That is why Gravitas Kitchen starts with inventory.</p><p>The inventory does not need to be perfect. It does not need expiration dates, exact quantities, or smart sensors. That kind of precision sounds good in theory, but it can become homework.</p><p>Instead, Gravitas Kitchen starts with a much lower-friction idea:</p><p>Tell the app what you have.</p><p>You can type it. You can dictate it. You can say something messy like:</p><p>“Bacon eggs cheese bread.”</p><p>Or:</p><p>“I’ve got chicken thighs, broccoli, rice, Greek yogurt, peanut butter, tortillas, salsa, some black beans, soy sauce, and green onions.”</p><p>That is a perfect use case for AI.</p><p>A human does not speak in database rows. A human speaks in fragments, pauses, corrections, grocery shorthand, and half-remembered items.</p><p>The AI’s job is to turn that unstructured speech into structured inventory.</p><p>Not by replacing the product with a brittle parser. Not by writing a bunch of comma-splitting rules and pretending that is intelligence.</p><p>The AI’s job is to understand the messy input and return clean structure:</p><pre>chicken thighs<br>broccoli<br>rice<br>Greek yogurt<br>peanut butter<br>tortillas<br>salsa<br>black beans<br>soy sauce<br>green onions</pre><p>That is the first step. It gives the kitchen memory.</p><h3>Have, Reorder, Need</h3><p>Once the app has memory, recipes become more useful.</p><p>In Gravitas Kitchen, ingredients are not just text. They become cards.</p><p>Each ingredient can be understood as one of four states:</p><p><strong>Have</strong> means the item is currently active in your inventory.</p><p><strong>Product</strong> means the app knows a more specific product match, like a certain yogurt, sauce, oat milk, or brand.</p><p><strong>Reorder</strong> means you do not currently have it, but it exists in your history. You bought it before, and it is probably something you may want again.</p><p><strong>Need</strong> means it is missing.</p><p>This is simple, but it changes how recipes feel.</p><p>A recipe no longer says:</p><p>“Here are nine ingredients. Good luck.”</p><p>It says:</p><p>“You have three of these. You have bought two before. You need four.”</p><p>That makes the recipe actionable.</p><p>This is where the agent idea starts to become real. The app is not just generating text. It is using memory to interpret the recipe in the context of your kitchen.</p><p>A recipe ingredient is no longer static. It has a relationship to your inventory, your history, and your shopping behavior.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o29NlaYcVYpZetgUG3zfSw.png" /></figure><h3>The Cookbook is taste memory</h3><p>The Cookbook is the second kind of memory.</p><p>Inventory remembers what you have.</p><p>The Cookbook remembers what you like.</p><p>You can save discovered recipes. You can write recipes manually. You can use AI Cleanup to turn a messy dictated recipe into something structured.</p><p>For example, you can type:</p><p>“Chicken broccoli rice thing. Chicken thighs, broccoli, rice, garlic, soy sauce, maybe green onions. Cook the rice. Cut up the chicken. Cook the chicken in a pan. Add broccoli and garlic. Add soy sauce. Serve over rice. Green onions if I have them.”</p><p>That is not a polished recipe. But it is exactly how people describe food in real life.</p><p>AI Cleanup turns that into a title, ingredients, optional ingredients, and ordered instructions.</p><p>The key is that the AI should preserve the user’s recipe. It should not invent a different dish. It should not add random ingredients. It should not turn a simple meal into a food blog production.</p><p>The app uses AI to structure the user’s intent, not to replace it.</p><p>That matters.</p><p>Because over time, the Cookbook becomes more than a recipe collection. It becomes a personal map of taste. It knows what you saved, what you cooked, what you can make again, and what ingredients connect to your kitchen memory.</p><h3>The Shopping List is not a list</h3><p>The most important design shift in Gravitas Kitchen might be the Shopping List.</p><p>In most apps, a shopping list is just text.</p><p>Milk.<br>Eggs.<br>Bread.<br>Soy sauce.</p><p>In Gravitas Kitchen, a Shopping List item is closer to a tiny search engine.</p><p>That sounds dramatic, but it is exactly how the grocery workflow works.</p><p>Take one item:</p><p><strong>Dole Whip Pineapple Ice Cream</strong></p><p>In a normal list, that is just a string.</p><p>In Gravitas Kitchen, that string can query a store’s product catalog. It can return candidate matches. It can store those results. It can let the user choose between them. It can remember the selected product. And then it can use the selected UPC to add the actual product to a cart.</p><p>So the item is no longer passive.</p><p>It has a query.<br>It has results.<br>It has a selected result.<br>It has a provider.<br>It has a store.<br>It has a fulfillment mode.<br>It has a possible UPC.<br>It has a path to action.</p><p>This is why the Shopping List is the decision surface.</p><p>A Shopping List item can search for itself.</p><p>For Kroger and Ralphs, the flow looks like this:</p><p>The user chooses a store.</p><p>The user chooses pickup or delivery.</p><p>Each Shopping List item searches for product matches at that store.</p><p>The app returns several product candidates.</p><p>The first result can be suggested automatically, but the user can open the item and choose another match.</p><p>Then the selected matches can be added to the user’s Kroger cart.</p><p>This turns “eggs” into an actual store product.</p><p>Not abstract eggs.<br>Not a random recipe ingredient.<br>Actual eggs at the selected store, with a product name, size, price if available, image, and UPC.</p><p>That is a very different kind of shopping list.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RZ1yzGJLAtzEXkA2L0A-Zw.png" /></figure><h3>Why this matters</h3><p>Food is one of the most repetitive decision systems in everyday life.</p><p>Every week, people ask the same questions:</p><p>What do we have?<br>What should we eat?<br>What do we need?<br>What should we buy?<br>Where should we buy it?<br>Can I get it delivered or pick it up?</p><p>Most apps only handle one piece of that.</p><p>A recipe app helps with ideas.</p><p>A grocery app helps with checkout.</p><p>A notes app holds a list.</p><p>A pantry app tracks inventory.</p><p>But the user still has to stitch the whole workflow together mentally.</p><p>Gravitas Kitchen tries to connect those pieces into one loop.</p><pre>Inventory<br>→ recipe ideas<br>→ ingredient cards<br>→ Shopping List<br>→ product matching<br>→ grocery cart<br>→ purchased items back to inventory</pre><p>That loop is why the app matters.</p><p>Not because it invents food planning from nothing.</p><p>Because it puts the pieces together in a way that reduces friction.</p><p>The magic is not that AI can produce a recipe. That is now ordinary.</p><p>The magic is that AI can sit inside a bounded workflow with memory and action.</p><h3>AI powers the messy parts</h3><p>There are parts of this app that should not be AI.</p><p>Adding a UPC to a Kroger cart should not be AI. That should be deterministic.</p><p>Matching an inventory item by UPC should not be AI. That should be deterministic.</p><p>Deduplicating a Shopping List item by normalized name should not be AI.</p><p>But the messy human parts are exactly where AI belongs.</p><p>Dictation is messy.</p><p>Recipe notes are messy.</p><p>Food preferences are messy.</p><p>Grocery intent is messy.</p><p>A human might say:</p><p>“No cilantro. I love garlic. Cheap meals. High protein if possible. Prefer larger recognizable brands. Avoid fancy ingredients I’ll only use once.”</p><p>That is not a form. It is not a dropdown. It is not a clean preference schema.</p><p>But it is useful.</p><p>The app can use that kind of preference text to influence recipe discovery and ranking.</p><p>A human might dictate an inventory list without commas.</p><p>“Bacon eggs cheese bread.”</p><p>The AI should know that those are separate grocery items.</p><p>A human might describe a recipe as a “thing.”</p><p>“Chicken broccoli rice thing.”</p><p>The AI should understand that the user is describing a simple dish, not asking for a gourmet reinvention.</p><p>That is the point.</p><p>The AI layer translates human mess into app structure.</p><p>Then the app uses deterministic systems to act on that structure.</p><p>That is the balance.</p><p>AI understands.<br>Software remembers.<br>APIs fulfill.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0UhfVLnXVHHmCE6h1A9LEQ.png" /></figure><h3>Why this is different from a recipe app</h3><p>Gravitas Kitchen should not be judged as a recipe app.</p><p>A recipe app starts with content.</p><p>Gravitas Kitchen starts with context.</p><p>The recipe is only useful after the app knows whether it fits your kitchen.</p><p>That is why the ingredient cards matter. That is why inventory matters. That is why the Shopping List matters. That is why product matching matters.</p><p>The app is not trying to answer:</p><p>“What recipes exist?”</p><p>It is trying to answer:</p><p>“What should I cook, given what I have, what I like, and what I can buy easily?”</p><p>That is a much more personal question.</p><p>And it is exactly the kind of question an AI agent should be able to help with.</p><h3>Why this is different from a grocery app</h3><p>A grocery app is built around a store.</p><p>Gravitas Kitchen is built around the household.</p><p>That distinction matters.</p><p>Kroger knows its products. Ralphs knows its stores. Instacart knows fulfillment.</p><p>But none of those services start with your kitchen memory.</p><p>They do not know that you already have rice.<br>They do not know that you bought soy sauce before.<br>They do not know that your Cookbook has three recipes using tortillas.<br>They do not know that “green onions” is optional for tonight.<br>They do not know that you are trying not to order takeout.</p><p>Gravitas Kitchen can sit above those systems.</p><p>It does not need to replace them. It should not replace them.</p><p>The grocery provider should still own checkout, payment, substitutions, pickup, and delivery.</p><p>Gravitas Kitchen owns the decision before the cart.</p><p>That is the layer that is missing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3oO6IHLUM61G1Cu6wLKBuw.png" /></figure><h3>Morning Brief gives it rhythm</h3><p>The Morning Food Brief gives Gravitas Kitchen a daily rhythm.</p><p>Instead of waiting for the user to start from zero, the app can say:</p><p>“Here are three meals you can make today.”</p><p>It can prefer Cookbook recipes first. It can include cook-again options. It can use current inventory. It can use Recipe Assistant when the Cookbook does not have enough good matches.</p><p>This is where the app starts to feel like an agent instead of a tool.</p><p>Not because it is autonomous in a dangerous way.</p><p>Because it has a rhythm.</p><p>It remembers, suggests, and prepares actions for the user to review.</p><p>That is the right amount of agency for food.</p><p>The app should not place an order automatically.</p><p>It should not decide dinner without you.</p><p>It should give you a head start.</p><h3>The loop after purchase</h3><p>One of the most important features is also one of the simplest.</p><p>After shopping, the user needs a low-friction way to move purchased items into inventory.</p><p>If the Shopping List helped create the cart, then the Shopping List should help update the kitchen memory afterward.</p><p>That is why a Shopping List item can be saved back into Inventory.</p><p>Swipe right.</p><p>Save to Inventory.</p><p>If the item was matched to a Kroger product, the product details come with it.</p><p>That closes the loop.</p><p>The item was needed.<br>It went onto the Shopping List.<br>It became a product match.<br>It went into the cart.<br>It was purchased.<br>It returns to inventory.</p><p>That is the product loop.</p><p>And once that loop works, the app becomes more useful every time you use it.</p><h3>Small app, complete loop</h3><p>Gravitas Kitchen is smaller than my other app, Gravitas Crunch.</p><p>But small does not mean insignificant.</p><p>In some ways, the smaller scope is its strength.</p><p>The app has a focused job:</p><p>Help you decide what to cook and what to buy.</p><p>That job is easy to understand. It is repeated often. It has a real-world outcome. And it benefits from AI without becoming vague.</p><p>The first version already has the core pieces:</p><p>Inventory.<br>Cookbook.<br>Recipe discovery.<br>Ingredient cards.<br>Shopping List.<br>Kroger and Ralphs product matching.<br>Cart handoff.<br>Morning Brief.<br>Save back to inventory.</p><p>That is not every grocery feature in the world.</p><p>But it is a complete loop.</p><p>And complete loops are where products become habits.</p><h3>The kitchen gets memory</h3><p>The phrase I keep coming back to is simple:</p><p><strong>Your kitchen gets memory.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XVa0jb6xbyj7h3n_SalJIQ.png" /></figure><p>That is what Gravitas Kitchen is really about.</p><p>It remembers what you have.</p><p>It remembers what you have bought.</p><p>It remembers recipes you care about.</p><p>It remembers what is missing.</p><p>It remembers product choices at the store.</p><p>And then it helps turn that memory into action.</p><p>Not an abstract AI chat.</p><p>Not another recipe feed.</p><p>Not another static shopping list.</p><p>A food agent with memory, structure, and a path to fulfillment.</p><p>That is why Gravitas Kitchen feels different.</p><p>It is not trying to make cooking magical.</p><p>It is trying to make cooking less mentally expensive.</p><p>And for most people, that is more useful.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b58db5051884" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Built a Personal Broadcast for the Real Web]]></title>
            <link>https://medium.com/@richfallatjrshop/i-built-a-personal-broadcast-for-the-real-web-07796d209d2f?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/07796d209d2f</guid>
            <category><![CDATA[apple]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[news]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Sun, 12 Apr 2026 03:26:35 GMT</pubDate>
            <atom:updated>2026-04-12T03:26:35.660Z</atom:updated>
            <content:encoded><![CDATA[<h3>Gravitas Crunch started with a simple feeling: I wanted to turn on the news, hear real reporting, and go straight to the source the moment something mattered.</h3><p>The killer feature of Gravitas Crunch is not a model name. It is not a benchmark. It is not even the interface, although I care deeply about that.</p><p>The killer feature is that I can just turn it on and get news on the hour.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wMPDYbPSWD796X6OjROMxg.png" /></figure><p>Not AI slop. Not synthetic filler. Real sources. Real publications. The kind of editorial you would actually trust with your attention. The kind of reporting that makes you stop what you are doing and tap <strong>Open Article</strong> because the imagery is too good, the story is too important, or the details are too rich to leave at the summary layer.</p><p>That single transition says almost everything about the app.</p><p>You hear the story.<br>You open the source.<br>You crunch it into understanding.<br>You ask questions.<br>You tap a citation and jump right back to the exact passage in the reader.</p><p>That is the loop.</p><p>What looks from the outside like a futuristic AI radio app is really the result of a much longer chain of ideas. Gravitas Crunch did not start as radio. It started as a reader, then kept pulling the next system into existence. Each piece created the need for the next one.</p><p>The story of the app is really the story of that chain.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3B97BP3nsYUjpmixzAGQGg.jpeg" /></figure><h3>It started with a reader</h3><p>I built Gravitas Crunch as a researcher.</p><p>That matters, because it shaped the product from the beginning. I did not want a toy summarizer. I wanted a place where I could actually live with information. I wanted to bring an article, a PDF, an EPUB, or a URL into an environment that respected the source, looked beautiful, and felt good enough to spend real time in.</p><p>The first truth I ran into was simple: a reader cannot be dead.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kUyVXvrHpsyjmkdB-8ahnA.png" /></figure><p>A lot of digital reading feels static and utilitarian. It gets the job done, but it does not feel like a medium. I wanted something with atmosphere. Something with motion. Something that felt closer to motion graphics than a white page with text dumped on it. I have always loved futuristic interfaces, especially the kind that feel cinematic rather than merely technical, and I wanted some of that energy here.</p><p>So the reader became more than a document view. It became a stage.</p><p>On some platforms, the article can auto-scroll so the page itself has movement. Images can rise into a lightbox carousel above the text so reading starts to feel a little like watching. There is a visualizer. There is glass. There is glow. There is a sense that the source is alive in front of you rather than frozen on the page.</p><p>That is where the project began: not with AI, but with the belief that source material deserved a better home.</p><h3>A reader is only as good as what it can open</h3><p>Once I had a reader, the next problem appeared immediately.</p><p>A beautiful reader is useless if it only works with a narrow, tidy set of inputs. Real research is messy. Real media is messy. People do not live in one file type or one publishing platform.</p><p>So the app had to learn how to open the world.</p><p>That meant resolvers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2jbB-NLpS0gyVyOjnOzqIA.png" /></figure><p>Gravitas Crunch can read and present PDFs, EPUBs, URLs, feeds, and a range of real publishing platforms. It grew resolvers for places like WordPress, Medium, Substack, arXiv, PubMed, DOI-based sources, code, and more. It can also resolve media from platforms like YouTube, Vimeo, and Reddit. Some sources need metadata extraction. Some need content cleanup. Some need platform-specific handling. Some need to respect premium access. Some are mostly text. Some are part text, part media, part structure.</p><p>That work was not glamorous, but it was foundational.</p><p>A source-aware app cannot treat the web like one giant undifferentiated blob. A paper is not a newsletter. A newsletter is not a blog post. A blog post is not a podcast episode. A podcast is not a PDF. Each one arrives with different structure, different context, and different expectations. The resolver layer was how all of those disparate things could be turned into a coherent internal model.</p><p>In other words, the reader forced me to build a language for content.</p><h3>Discovery had to be part of the product, not an afterthought</h3><p>Once the app could open sources well, another question appeared: how do those sources get there?</p><p>I did not want discovery to feel generic. I wanted it to feel editorial.</p><p>So I built discovery around a search system that uses Google as the backbone, but routes through tailored engines for different content types. Feed discovery behaves differently than EPUB discovery. DOI is not the same as PubMed. Medium is not the same as Substack. Code is not the same as a news article. Each engine is trying to find the right kind of thing, not just anything that matches a string.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CeVxcTChAsth_QriniDsFw.png" /></figure><p>That matters because discovery shapes the quality of the whole experience.</p><p>If the inputs are low quality, the outputs are low quality. If the inputs are generic, the whole product starts to feel generic. But if discovery is source-aware and editorially grounded, then the rest of the system has real material to work with.</p><p>This is one of the reasons the app feels different to me. It is not generating a fake world. It is curating access to the real one.</p><p>That distinction matters more than almost anything else.</p><h3>Then came the real problem: there is too much worth reading</h3><p>The web does not have an information shortage. It has a backlog problem.</p><p>There is far more worth reading than most people have time to read. Articles are long. Papers are dense. Books are huge. Even when the source is excellent, the cost of keeping up can be too high.</p><p>That is where <strong>Crunch</strong> came from.</p><p>Once you open a source in Gravitas Crunch, the major call to action is not “read more.” It is <strong>Crunch</strong>. There is even a little subtitle beneath it: <em>Don’t just read it. Crunch it.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P8HuS-jBvl3LmzpDa2B_Mw.png" /></figure><p>That line gets at the heart of what I wanted the app to do.</p><p>Crunch is not a one-line summary. It is not a lazy paraphrase. It is the system’s comprehension engine. Long-form content is broken into chunks, summarized at the chunk level, and assembled into a structured outline. That outline becomes a working representation of the source: something far more usable than the raw document alone, but still grounded enough to preserve its shape and meaning.</p><p>I designed multiple ways to crunch content because different reading goals require different compression. Sometimes you want an executive view. Sometimes you want a technical lens. Sometimes you want a study guide. And sometimes you want what I call <strong>Focus</strong>: a question-directed pass over the full source where the summarization itself is guided by what you want to know.</p><p>That last part is important.</p><p>A lot of summarization is neutral by default. It reduces everything evenly. But real questions are not neutral. If you feed the system a novel and ask about a specific character’s journey, the goal should not be to flatten the whole book into a general-purpose recap. The goal should be to compress the book in a way that serves that question.</p><p>That is what Focus is for.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dru03Rcba8YJEm30FY3SzA.png" /></figure><h3>Research was the natural next step</h3><p>Once Crunch existed, the app stopped being just a reader.</p><p>It became a research assistant.</p><p>The moment a source has been crunched into a structured outline, you no longer have to treat it as a wall of text. You can read the outline directly. You can use it as a quick way back into the source. You can work from it. You can chat against it. You can use it to generate something larger and more specific. You can even export notes to PDF.</p><p>This was a big shift for me.</p><p>I did not want research to feel like throwing a document into a black box and getting back an answer you could not inspect. I wanted it to feel like turning a source into a usable knowledge object. Something you could return to. Something you could probe. Something that gave you leverage over long-form content instead of simply reducing it.</p><p>That is what Crunch really became: not just a summary action, but a compilation step for understanding.</p><h3>Answers had to show their work</h3><p>The next problem was trust.</p><p>The moment you add an assistant layer to content, you risk floating away from the source. A lot of AI experiences feel that way to me. They are smooth, but untethered. They talk confidently, but the underlying material is hidden.</p><p>I did not want that.</p><p>So when the research assistant returns an answer, it can also return citation buttons. Tap one, and you jump directly to the exact spot in the reader that supports the point.</p><p>That feature changed the feel of the whole product.</p><p>Suddenly the assistant was not speaking in the air. It was rooted in the source. The source was still primary. The assistant was a guide, not a replacement. It could help you move faster, but it could also bring you right back to the text, the passage, the evidence.</p><p>That is the line I cared about preserving.</p><p>Gravitas Crunch does not hide the source. It brings you back to it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K-i435s06hVkd8Gju9y4Gw.png" /></figure><h3>Radio was not the beginning. It was the inheritance of everything before it</h3><p>People tend to understand the app first through <strong>Robot Radio</strong>, because it is easy to grasp. Turn your feeds into a station. Press play. Let it run.</p><p>But the reason Robot Radio works is that all the earlier systems already existed.</p><p>The app already knew how to open sources.<br>It already knew how to normalize them.<br>It already knew how to chunk them.<br>It already knew how to turn them into structured understanding.<br>It already knew how to preserve a path back to the source.</p><p>Radio emerged when I realized all of that understanding could be turned into a listening medium.</p><p>That required another transformation layer: <strong>Anchor Mode</strong>.</p><p>Anchor Mode takes the structured understanding of a source and rewrites it for the ear. It is not enough to simply read article prose aloud. Written prose and spoken presentation are not the same medium. So the app packages the story like a broadcast segment. It shapes utterances for delivery. It rewrites in a news-anchor style. It spells out acronyms and abbreviations so they sound right when spoken. It uses Apple Intelligence to segment text into more natural utterances before synthesis. And it can choose bumper music based on the tone and type of the story from a library of authored tracks tagged with metadata for mood and use.</p><p>That is an important design philosophy for me.</p><p>I did not want the model inventing the whole experience from scratch. I wanted it orchestrating a designed system: choosing, packaging, and composing inside constraints that I authored. The music is curated. The metadata is curated. The broadcast rules are curated. The model’s job is judgment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t8u6vR2UU0Yg1ZaWbLSjyg.png" /></figure><p>That is what makes the segments feel cohesive.</p><p>And then there was the engineering reality behind it.</p><p>Text-to-speech is computationally expensive. High-quality playback cannot stutter every time a story changes. Some parts of the pipeline can run in parallel, but others cannot. So I had to build a compute-ahead rendering pipeline that could stay ahead of playback, time pauses correctly, and keep the station feeling seamless. I also had to be disciplined enough to let very long items go when necessary so one giant source would not hijack the listening experience.</p><p>That was brutal work. But it was worth it, because once it came together, the app crossed a threshold.</p><p>It stopped feeling like a feature.<br>It started feeling like a medium.</p><h3>Audiobook mode took the same foundation in a different direction</h3><p>Radio is discovery. Audiobook mode is depth.</p><p>Once a story catches your attention and you open the source, there are times when you do not want a radio segment. You want the full thing. You want the article, the essay, the paper, the EPUB, the long-form piece, and you want it narrated in a way that lets you stay with it.</p><p>That required a different orchestration pipeline.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yKMTsLl9eZGwloXsLN4uew.png" /></figure><p>Radio has to move. It has to keep the broadcast alive. Audiobook mode can linger. It can honor the full structure of the source. It can live much closer to the original while still benefiting from all the work that went into utterance quality, pacing, and presentation.</p><p>I love that the same source can move through both modes.</p><p>You can hear the high-level segment first.<br>Then you can open the article.<br>Then you can switch into a deeper listening mode and stay with it longer.</p><p>That is the kind of continuity I wanted the product to have all along.</p><h3>A station is only as good as its editorial</h3><p>None of the radio work would matter without the feed network.</p><p>Gravitas Crunch has a curated library of feeds across news, politics, business, tech, science, space, gaming, entertainment, beauty, DIY, sports, real estate, health, justice, and more. The point is not sheer quantity for its own sake. The point is that the editorial quality is high enough to sustain listening.</p><p>That is what made me realize the product had something real.</p><p>I could leave it on and actually want to keep listening.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DgEQ3bPdtFBhIX6lYHvLVg.png" /></figure><p>The app can create different station experiences from that network. There is <strong>Now Radio</strong>, which sweeps through the latest items from the broader library and your own feeds. There is <strong>My Station</strong>, which is where the product becomes truly personal: you can follow feeds, reorder the broadcast, shuffle, go back through history, choose exactly what participates, and shape the station the way you want it.</p><p>That is when the app stopped feeling like an app and started feeling like a personal channel.</p><p>And the beautiful part is that this works at both a broad and intimate scale. One day it can be a constant stream of world news. Another day it can be a station built around a narrow interest. It can even become a custom editorial space for another person entirely.</p><h3>The real destination is a personal broadcast</h3><p>This is the part I love most.</p><p>At the end of the chain, Gravitas Crunch is not just a reader, not just a research assistant, and not just a radio app. It is a way of turning trusted sources into a personal broadcast.</p><p>That phrase gets at something deeper than feature count.</p><p>A personal broadcast is not the same thing as a recommendation feed. It is not an algorithm guessing what will keep you scrolling. It is something more intentional. It is built from the publications you trust, the subjects you care about, the sources you follow, the research you save, the way you like to listen, and the questions you bring to long-form material.</p><p>It is a medium shaped around your mind.</p><p>That is why the product feels so different to me from generic AI wrappers. It is not generating substitute content. It is creating a new way to move through real content.</p><h3>I designed each Apple platform as its own experience</h3><p>Because the medium changes, the form factor has to change too.</p><p>Gravitas Crunch is multiplatform, but I never wanted that to mean one interface stretched across every device. Each platform got its own care.</p><p>On <strong>iPhone</strong>, the app feels intimate and immediate. It is the quickest way to turn on the station, open a story, and keep moving.</p><p>On <strong>iPad</strong>, the reader has room to breathe. It becomes a more relaxed canvas for reading, browsing, and shifting between listening and source work.</p><p>On <strong>macOS</strong>, the product becomes a workstation. Discovery, source reading, and research can live together on screen in a way that makes the whole architecture visible at once.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S9vpn1VawgnvfJYcQSnlvQ.jpeg" /></figure><p>On <strong>Apple Vision Pro</strong>, the spatial and cinematic side of the app becomes especially meaningful. Images, motion, article surfaces, and atmosphere all have room to become part of the experience instead of just the frame around it.</p><p>Across the platforms, I used Apple’s Liquid Glass as a starting point, but pushed it into a softer frosted-glass look that felt more like light diffusing through a material than hard glossy shine. I borrowed from my background in graphics and from the motion design I love: subtle refraction, layered glow, auto-scrolling surfaces, a visualizer that reflects the inner life of the pipeline, and an overall sense that the app is not merely presenting information but staging it.</p><p>I wanted every platform to feel like the same world, expressed differently.</p><h3>What I really built</h3><p>From the outside, the easiest way to describe Gravitas Crunch is still Robot Radio. That is the hook. That is the thing people immediately understand.</p><p>But the truth is that the app is a chain of systems built on top of a single idea:</p><p>real sources deserve a better way to be found, read, understood, heard, and returned to.</p><p>That idea became a reader.<br>The reader became resolvers.<br>Resolvers forced discovery.<br>Discovery demanded better summarization.<br>Summarization became research.<br>Research required citations.<br>Citations kept the source alive.<br>All of that became radio.<br>Radio expanded into audiobook mode.<br>The feed network became stations.<br>The stations became a personal broadcast.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dfQ5sECfnvw4c3BkFfB1lA.jpeg" /></figure><p>So yes, the killer feature is that I can turn it on and get news on the hour.</p><p>But the reason that feature feels magical is everything underneath it.</p><p>Gravitas Crunch is my attempt to turn the web, documents, and long-form media into something you can actually live with: a reader when you need source truth, a research assistant when you need understanding, and a broadcast when you simply want the world to start talking.</p><p>That is the product.</p><p>Hear the story.<br>Open the source.<br>Crunch it.<br>Ask it.<br>Verify it.<br>Keep going.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=07796d209d2f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Turn the Sources You Follow Into Your Own Personal Radio Station]]></title>
            <link>https://medium.com/@richfallatjrshop/turn-the-sources-you-follow-into-your-own-personal-radio-station-679de76b9471?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/679de76b9471</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[ios]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Tue, 10 Mar 2026 12:34:12 GMT</pubDate>
            <atom:updated>2026-03-11T03:17:32.077Z</atom:updated>
            <content:encoded><![CDATA[<p>The internet was supposed to make it easier to stay informed.</p><p>Instead, for a lot of us, it became homework.</p><p>Too many tabs. Too many saved links. Too many PDFs. Too many newsletters, feeds, articles, videos, and things we meant to come back to later. Information is everywhere, but actually keeping up with it often feels fragmented, tiring, and high-friction.</p><p>That’s the problem I’ve been trying to solve with <strong>Gravitas Crunch</strong> for iPhone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c_Ou2oEbeEc6U52itkx0fA.png" /></figure><p>Gravitas Crunch is a private research workspace built around a simple idea:</p><p><strong>Start with real sources. Turn long reads into clear takeaways. Drill back into the original source when something matters.</strong></p><p>And most importantly:</p><p><strong>Turn the sources you follow into your own personal radio station.</strong></p><h3>The problem with modern information overload</h3><p>A lot of information tools still force the same pattern:</p><p>You find something interesting.<br> You save it.<br> You open it later.<br> You skim it halfway.<br> You switch apps.<br> You lose the thread.<br> You forget where the useful part was.<br> You move on.</p><p>That cycle is incredibly common now. It happens across browser tabs, read-later apps, PDFs, notes, feeds, and social links. Even when the source material is high quality, the workflow around it often isn’t.</p><p>For me, feeds were part of that problem.</p><p>I understood why they were useful in theory, but they still felt like an old internet artifact. Something powerful, but also technical, manual, and oddly distant from how most people actually consume information now.</p><p>What changed for me was rethinking feeds not as a reader problem, but as a <strong>listening</strong> problem.</p><p>What if the sources you follow could come to you?</p><p>What if they could feel less like a backlog and more like a station?</p><p>That question led to <strong>Gravitas Radio</strong> inside Gravitas Crunch.</p><h3>What Gravitas Crunch does</h3><p>Gravitas Crunch is designed to help you move from information overload to understanding with less friction.</p><p>It includes:</p><ul><li><strong>Gravitas Radio</strong> for continuous listening from the sources you follow</li><li><strong>Crunch</strong> modes that turn long reads into clear takeaways</li><li><strong>Source-grounded drilldown</strong> so you can jump back into the original material</li><li><strong>Source discovery</strong> for finding real articles and feeds quickly</li><li><strong>Import support</strong> for URLs, PDFs, EPUBs, feeds, and more</li></ul><p>The goal is not generic AI chat.</p><p>The goal is a workflow that stays connected to the original source material while making it easier to absorb, revisit, and follow up on what matters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p_P6QFAvPsnwv1RM4y7BFA.png" /></figure><h3>Gravitas Radio: from feeds to personal broadcast</h3><p>The most distinctive part of the app is <strong>Gravitas Radio</strong>.</p><p>Instead of treating the things you follow as a static reading list, Gravitas Radio turns them into a continuous listening experience. Articles from your followed sources can be distilled into clear narration and played back like a personal station.</p><p>That changes the feeling of the whole workflow.</p><p>Instead of asking:</p><p>“What should I read next?”</p><p>the app can move toward:</p><p>“What do I want to listen to, and what do I want to go deeper on?”</p><p>That difference matters.</p><p>Listening is lower-friction. It fits more moments in a day. It works while walking, cleaning up, getting ready, commuting, or stepping away from a screen. It lets high-level understanding come first, and deeper reading come second.</p><p>For a lot of people, that is a much more natural way to keep up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zImkuyL74VBifZHbDqQd4w.png" /></figure><h3>Crunch: long reads into takeaways</h3><p>Not every article needs full attention up front.</p><p>Some things deserve a quick high-level pass first. Others deserve more detailed analysis.</p><p>That’s where <strong>Crunch</strong> comes in.</p><p>Gravitas Crunch can help turn long-form sources into clearer takeaways so you can understand the shape of a piece before deciding how deep to go.</p><p>This is especially useful when dealing with:</p><ul><li>dense articles</li><li>technical posts</li><li>long-form essays</li><li>saved reading backlogs</li><li>PDFs and papers</li><li>sources you want to understand quickly before committing more time</li></ul><p>The point is not to replace reading.</p><p>The point is to make reading more intentional.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*da-ZribynmB1MpXaG7yEfg.png" /></figure><h3>Source-grounded, not detached</h3><p>One of the biggest problems with a lot of AI-driven tools is that they drift away from the source.</p><p>You get an answer, but it becomes harder to tell where it came from. The convenience is high, but the trust can be low.</p><p>Gravitas Crunch is built around the opposite principle:</p><p><strong>See the source behind the answer.</strong></p><p>That means the workflow does not end at a summary or generated narration. If something is important, you should be able to drill back into the original source and keep exploring in context.</p><p>That makes the app more useful for real research, not just quick novelty.</p><h3>Find real sources fast</h3><p>Before you can read, listen, or research, you need good input.</p><p>That is why Gravitas Crunch also includes source discovery. The goal is to make it easier to find real material quickly, whether that means articles, feeds, publications, or other long-form sources worth following.</p><p>This matters because information quality often starts with where you are sourcing from in the first place.</p><p>If the inputs are better, the downstream experience becomes better too.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rvw8ja63R0efpOMVX6qNEw.png" /></figure><h3>Import articles, feeds, PDFs, EPUBs, and more</h3><p>Gravitas Crunch is not meant to be limited to one format or one style of source.</p><p>Information lives in a lot of places now. Useful material shows up as links, feeds, PDFs, EPUBs, articles, and other formats. A modern research workflow should not force those into completely separate worlds.</p><p>That is why import matters.</p><p>The more of your real information life the app can handle, the more valuable the experience becomes.</p><h3>Why this approach matters</h3><p>The biggest idea behind Gravitas Crunch is simple:</p><p><strong>People do not just need more access to information. They need better ways to absorb it.</strong></p><p>Feeds alone are not enough. Saving things for later is not enough. Summaries alone are not enough either.</p><p>What matters is the full path:</p><ul><li>find something real</li><li>understand it at a high level</li><li>listen when reading is too much friction</li><li>go deeper when something matters</li><li>stay connected to the source</li></ul><p>That is the workflow I wanted for myself, and that is what Gravitas Crunch is trying to become.</p><h3>A more useful role for AI</h3><p>There is a lot of noise around AI right now.</p><p>A lot of products lead with the model instead of the workflow. A lot of features feel impressive for a minute but do not become part of real daily use.</p><p>What interests me more is using these tools in a way that makes existing behaviors better.</p><p>In this case, that means helping real sources become:</p><ul><li>easier to keep up with</li><li>easier to listen to</li><li>easier to revisit</li><li>easier to trust</li></ul><p>That is a much more grounded role.</p><h3>The bigger shift</h3><p>I think a lot of internet products still assume the user wants to sit down and manually pull information toward themselves all day.</p><p>But increasingly, the better experience may be the opposite:</p><p>Let the important things come to you in a cleaner, calmer, more usable form.</p><p>That is what makes the idea of a personal source-driven radio station interesting to me. It takes something that used to feel like maintenance and turns it into something that feels much more natural.</p><p>Feeds stop being a chore.<br> Reading backlog stops being guilt.<br> Research becomes less fragmented.<br> The internet becomes a little less like homework.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AmG6b59WZ-4gwJTdE401fA.png" /></figure><h3>Final thought</h3><p>Gravitas Crunch is still built around a straightforward belief:</p><p><strong>Real sources matter.</strong><br> <strong>Good workflows matter.</strong><br> <strong>And staying informed should not feel this hard.</strong></p><p>If we can turn the sources we care about into something we can actually live with day to day, that alone is valuable.</p><p>That is the promise behind Gravitas Crunch.</p><p><strong>Turn the sources you follow into your own personal radio station.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=679de76b9471" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Real Debate About “Vibe Coding” Has Nothing to Do With Perfection]]></title>
            <link>https://medium.com/@richfallatjrshop/the-real-debate-about-vibe-coding-has-nothing-to-do-with-perfection-f25c15e484e1?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/f25c15e484e1</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Mon, 09 Mar 2026 05:49:32 GMT</pubDate>
            <atom:updated>2026-03-09T05:52:06.535Z</atom:updated>
            <content:encoded><![CDATA[<p>AI-assisted development is not about avoiding work. It is about removing execution as the bottleneck between ideas and software.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SG_2OruYpK90OpeOZgsSKA.png" /><figcaption>Generated with ChatGPT 5.2</figcaption></figure><p>There’s a phrase people use now — <strong>“vibe coding”</strong> — and most of the time it isn’t descriptive.</p><p>It’s dismissive.</p><p>It’s a way to wave away a real shift in software creation by making it sound unserious, sloppy, or fake. As if using AI to build means you’re just improvising your way through code and hoping it somehow works.</p><p>That is not what this is.</p><p>For me, AI-assisted development is not about being lazy. It is about removing friction between <strong>ideas</strong> and <strong>execution</strong>.</p><p>My workflow is simple.</p><p>I use <strong>ChatGPT Pro</strong> as the strategy layer. That is where I think through architecture, research, platform constraints, implementation plans, and the cleanest way to structure a system. Then I take those instructions into <strong>VS Code</strong> and hand them to <strong>GPT-5.2</strong>, which becomes the execution layer. It writes and refactors the code based on a plan that is already far more comprehensive than what I would have produced alone in the same amount of time.</p><p>Very occasionally, when I need to bring code back into ChatGPT Pro for deeper reasoning, I export files as text and round-trip them back in. That is rare, but it happens.</p><p>What matters is the handoff.</p><p>I’m not manually translating every thought into syntax anymore. I’m letting the models talk to each other in a level of detail that would take me much longer to express by hand.</p><p>And the result, honestly, feels like a <strong>100,000x speedup</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NvVG0PNlBxHfmvgzXmts-Q.png" /><figcaption>Generative AI On-device Multimodal Story Studio built with ChatGPT</figcaption></figure><p>Not as a benchmark. As a lived experience.</p><p>The difference is so dramatic that it breaks the old mental model of what software creation feels like.</p><p>I recently ported a fairly complicated reader app from <strong>Apple Vision Pro</strong> to <strong>iPhone</strong> in under 24 hours. This was not a toy project. It was around <strong>10,000 lines of code</strong>, with lots of resolver logic. In another era, that kind of migration would have felt like a heavy lift. With this workflow, it felt possible, fast, and creatively energizing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c_Ou2oEbeEc6U52itkx0fA.png" /><figcaption>Apple iOS AI powered Reader and Feed Radio app built in under a week.</figcaption></figure><p>That’s why I think the debate around “vibe coding” misses the point.</p><p>The most common criticism I hear is always some version of this:</p><p><strong>“Sure, but it’s not perfect.”</strong></p><p>Okay.</p><p>And what is?</p><p>Human-written code is not perfect either.</p><p>One of the strangest standards people apply to AI-generated code is that it must somehow be <em>better than human code to count at all</em>. It has to be elegant, fully optimized, bug-free, architecturally pristine, and production-perfect on arrival, or else people dismiss it.</p><p>That standard is fantasy.</p><p>Most code is compromise.</p><p>Most products are compromise.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e5mjeSpU9mbcm8K3Iza8pA.png" /><figcaption>Apple Vision Pro Mulimodal AI Story Studio built with ChatGPT</figcaption></figure><p>Software gets built under constraints: time, money, deadlines, changing requirements, incomplete information, skill gaps, and plain old human fatigue. We do not live in a world of perfect systems. We live in a world of trade-offs.</p><p>Yes, there are domains where the bar has to be much higher. Safety-critical systems, certain financial systems, certain infrastructure layers — those deserve tighter guarantees. But that does not make “it’s not perfect” a serious critique of most real-world product development.</p><p>The real question is not:</p><p><strong>Is this code perfect?</strong></p><p>The real question is:</p><p><strong>Does it perform the way I need it to perform?</strong></p><p>If it gets me 90% or 95% of the way there, and the remaining work is manageable, that is not a failure.</p><p>That is leverage.</p><p>And over the last year and a half of working this way, I have not hit a hard wall yet.</p><p>I have had moments where the output wasn’t good enough, so I passed.</p><p>I have had moments where it was too expensive, so I passed.</p><p>I’m not claiming AI is magic in every situation.</p><p>But I have not hit the wall that critics keep insisting must be there.</p><p>I stumble sometimes.</p><p>I have not fallen.</p><p>Ironically, when I code everything myself, I hit walls all the time. I can lose an entire day chasing one bug. I can get trapped in implementation details. I can spend hours being slow in ways that only feel normal because we were trained to accept that slowness as the price of creation.</p><p>That is the deeper shift.</p><p>My goal was never to type code.</p><p><strong>Typing is not the product. The product is the product.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8IA-Ynuw31u1hfwIZ0e_yA.png" /><figcaption>Generated with ChatGPT 5.2</figcaption></figure><p>If a machine can handle syntax, boilerplate, recall, framework trivia, and a huge amount of implementation detail, then I would rather spend my time directing, testing, judging, refining, and shipping.</p><p>That is not cheating.</p><p>That is using tools correctly.</p><p>And this is where the phrase <strong>“vibe coding”</strong> becomes revealing.</p><p>Because what AI actually unlocks is not lower standards. It unlocks <strong>access</strong>.</p><p>There are a lot of people with strong product instincts, strong ideas, strong taste, and strong ambition who were previously blocked by execution. Not everyone has the time — or the desire — to become deeply expert in every framework, every language, every platform nuance, every optimization strategy.</p><p>And now, suddenly, that bottleneck is weaker.</p><p>That is not a decline.</p><p>That is an expansion of who gets to build.</p><p>So when someone mocks “vibe coding,” what I often hear underneath it is discomfort with the fact that the gate is lower now. Syntax memorization matters less. Typing speed matters less. Knowing obscure implementation details off the top of your head matters less.</p><p>Judgment matters more.</p><p>Taste matters more.</p><p>Direction matters more.</p><p>The bottleneck is moving from <strong>syntax</strong> to <strong>decision-making</strong>.</p><p>That is a big deal.</p><p>And it is good.</p><p>Because the people using these tools well are often not people with less drive. They are people with more ideas than time. People who were blocked from execution. People who can now finally turn thought into software at something closer to the speed of imagination.</p><p>That is what AI changes.</p><p>It does not remove responsibility. You still need to review what gets built. You still need standards. You still need testing. You still need accountability for whether the software works, whether it performs, and whether it delivers value.</p><p>But it radically changes the cost of getting from <strong>idea</strong> to <strong>implementation</strong>.</p><p>And that matters more than people want to admit.</p><p>So no, I don’t really take “vibe coding” as an insult.</p><p>I hear it as the language people use during a transition, when they are trying to describe a new way of building with the assumptions of the old one.</p><p>Call it whatever you want.</p><p>I call it leverage.</p><p>I call it accessibility.</p><p>I call it software creation becoming more human, not less — because humans get to spend less time typing and more time deciding what should exist.</p><p>And if your only argument against that is that the code is not perfect, then you are defending a fantasy that never existed in the first place.</p><p>Maybe the better question is not whether AI-assisted coding looks pure enough to impress other programmers.</p><p>Maybe the better question is much simpler:</p><p><strong>Did you build something that works?</strong></p><p>If the answer is yes, the debate is already moving on without you.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/1*8NTiJSu6xV-bebcaxhqmYQ.png" /><figcaption>Apple Apps created with ChatGPT</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f25c15e484e1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gravitas Reader turns AR into a multi-room content magazine — now let’s look under the hood]]></title>
            <link>https://medium.com/antaeus-ar/gravitas-reader-turns-ar-into-a-multi-room-content-magazine-now-lets-look-under-the-hood-9ab6318f4bbb?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/9ab6318f4bbb</guid>
            <category><![CDATA[augmented-reality]]></category>
            <category><![CDATA[apple-vision-pro]]></category>
            <category><![CDATA[virtual-reality]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Sat, 08 Nov 2025 08:37:25 GMT</pubDate>
            <atom:updated>2025-12-10T05:46:18.670Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Gravitas Reader turns AR into a multi-room content magazine — now let’s look under the hood</strong></h3><p>Gravitas Reader explores what reading might feel like in augmented reality.<br> It’s built for Apple Vision Pro and takes familiar things — articles, PDFs, and videos — and lets them live in space instead of on flat screens.</p><p>This piece looks at how the app works behind the scenes: how it displays articles as spatial layouts, connects to real sites like Medium, and lets readers keep what they find in a way that feels natural.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s910GqoqZpxipf4JW2_IuA.jpeg" /><figcaption><strong>Reading Arc</strong></figcaption></figure><h3>Spatial reading as a layout engine</h3><p>At the center of the app is something we call the curved reading wall. It’s a wide arc that holds several pieces of a story at once — headlines, images, and key quotes. The goal isn’t to impress with design; it’s simply to make long-form reading easier to follow when your eyes can wander freely.</p><p>Inside the system, each article is broken down into small, predictable blocks — paragraphs, lists, or images — and then laid out across that wall. That simple structure keeps things consistent, no matter if you’re looking at a PDF, a blog post, or a forum thread.</p><pre>struct ArticleBlock: Codable {<br>    enum Kind: String, Codable { case heading, paragraph, listItem, image, video, code }<br>    let kind: Kind<br>    let text: String?<br>    let imageURL: URL?<br>    let videoURL: URL?<br>    let caption: String?<br>    let codeLanguage: String?<br>}</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*grWmUkBBi0xRBpN6M_mcWQ.jpeg" /><figcaption><strong>Card Detail</strong></figcaption></figure><h3>Source-specific resolvers + a shared extractor</h3><p>Every website structures its content differently. To handle that, we built small modules — called resolvers — that know how to extract clean text and media from a handful of popular sources like Medium, Reddit, and YouTube. When a site doesn’t match any resolver, the app just loads it in a lightweight web view and gathers what it can.</p><p>This approach isn’t perfect, but it’s flexible. It keeps the layout steady and gives readers a chance to view almost any source without switching context.</p><pre>protocol URLResolver {<br>    func canResolve(_ url: URL) -&gt; Bool<br>    func resolve(_ url: URL, signer: RequestSigner) async throws -&gt; Article?<br>}<br><br>struct MediumResolver: URLResolver {<br>    func resolve(_ url: URL, signer: RequestSigner) async throws -&gt; Article? {<br>        var request = URLRequest(url: apiURL(for: url))<br>        request.setValue(mediumUserAgentString, forHTTPHeaderField: &quot;User-Agent&quot;)<br>        signer.sign(&amp;request)<br><br>        let (data, response) = try await URLSession.shared.data(for: request)<br>        try PaywallDetector.throwIfPaywalled(data: data, response: response, url: url)<br>        return try decodeArticle(from: data, sourceURL: url)<br>    }<br>}</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HzAQXN3frdPgfHzm4eHY5A.jpeg" /><figcaption>YouTube support</figcaption></figure><h3>“Connect” for premium content</h3><p>Some sites, like Medium, require login access. To make that smoother, we added a way for readers to connect their accounts securely. Cookies are stored on the device, not on our servers, and once you link a site, you don’t need to keep logging in.</p><pre>struct RequestSigner {<br>    let apply: (inout URLRequest) -&gt; Void<br>    static let none = RequestSigner { _ in }<br>}<br><br>enum PaywallDetector {<br>    static func throwIfPaywalled(data: Data, response: URLResponse, url: URL) throws {<br>        guard let http = response as? HTTPURLResponse else { return }<br>        if [401, 402, 403, 451].contains(http.statusCode) {<br>            throw Error.paywalled(site: url.host ?? &quot;site&quot;)<br>        }<br>        let snippet = String(data: data.prefix(2048), encoding: .utf8)?.lowercased() ?? &quot;&quot;<br>        if [&quot;subscribe&quot;, &quot;subscriber-only&quot;, &quot;paywall&quot;].contains(where: snippet.contains) {<br>            throw Error.paywalled(site: url.host ?? &quot;site&quot;)<br>        }<br>    }<br>}</pre><p>If a story is behind a paywall, the app simply shows a “Connect to continue” card. It’s a small detail, but it helps keep the reading flow from breaking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZTB9i6BSk-tV_41UcWDySA.jpeg" /><figcaption><strong>Medium Auth Flow</strong></figcaption></figure><h3>Museum Mode + Scenes</h3><p>One of the more personal features is called Museum Mode. It lets readers pin quotes, images, or charts onto their walls as visual notes — things they might want to remember or think about later.</p><p>Another feature, Scenes, saves your whole reading setup: where your articles are placed, what you’ve pinned, and how you’ve arranged the space. When reopened, it’s like walking back into a room you were studying in earlier.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*plZvKHgLlTlzx4M5r-7eFQ.jpeg" /><figcaption><strong>Museum / Scene Showcase</strong></figcaption></figure><h3>Discover Assistant + Research Assistant</h3><p>Gravitas Reader isn’t just a wall of text; it has two helpers built in.</p><ul><li><strong>Discover Assistant</strong> — It’s a Google-powered search and discovery layer. You type a query, it searches the web, and then pulls clean previews through the same resolvers used for layout (Medium, PDFs, Reddit, etc.), so results render consistently before you open them.</li><li><strong>Research Assistant</strong> — Once you’re deep in a long article or book, this AI mode analyzes the article (including books), generates concise summaries, and keeps provenance so you can jump straight to the cited page. Executive summaries run once per session, while interactive chat lets you ask questions about the current article without leaving the spatial layout. It keeps the context of the article during the conversation—chatGPT for example does not keep articles in memory.</li></ul><p>Together, the two assistants surface what to read next, then help you stay oriented inside dense material — without breaking the spatial reading flow.</p><p>Foundation model example:</p><pre>// ResearchAssistantService.swift<br>private func generateReply(for userText: String) async throws -&gt; String {<br> // 1) Create a Foundation session from our pre-baked system prompt.<br> guard let chat = Self.makeFoundationSession(system: instructions) else {<br> throw ResearchAssistantError.modelUnavailable<br> }<br>// 2) Budget: 1,250 tokens total for chat → per-chunk token cap.<br> let aggregateBudgetTokens = 1_250<br> let perChunkTokens = Int(ceil(Double(aggregateBudgetTokens) /<br> Double(max(convo.summaryChunks.count, 1))))<br> let perChunkChars = perChunkTokens * 4 // rough token→char map<br>var workingChunks = convo.summaryChunks<br> var didClamp = false, didConcat = false<br> var attempt = 1<br>while true {<br> // 3) Build prompt (Reference data + Question).<br> let (prompt, tokenEstimate) = try buildPrompt(for: userText, chunks: workingChunks)<br> GRLog.p(&quot;ResearchAssistant&quot;, &quot;[chat-prompt] size: ~\(tokenEstimate) tokens (\(prompt.count) chars)&quot;)<br>do {<br> // 4) Ask the model.<br> GRLog.p(&quot;ResearchAssistant&quot;, &quot;[chat] awaiting model response (attempt \(attempt))…&quot;)<br> let response = try await chat.respond(to: prompt)<br> let final = response.content.trimmingCharacters(in: .whitespacesAndNewlines)<br> guard !final.isEmpty else { throw ResearchAssistantError.emptyResponse }<br> return final<br> } catch {<br> // 5) Fail-safes: clamp → concat → give up.<br> GRLog.p(&quot;ResearchAssistant&quot;, &quot;[chat] LLM failed on attempt \(attempt): \(error.localizedDescription)&quot;)<br> if !didClamp {<br> didClamp = true<br> workingChunks = ResearchSummaryService.shared.clampAggregate(<br> workingChunks, maxCharacters: perChunkChars<br> )<br> GRLog.p(&quot;ResearchAssistant&quot;, &quot;[chat] clamped over-budget chunk summaries to \(perChunkChars) chars; retrying…&quot;)<br> attempt += 1<br> continue<br> } else if !didConcat {<br> didConcat = true<br> workingChunks = ResearchSummaryService.shared.concatenateAggregate(<br> workingChunks,<br> maxCharacters: perChunkChars * max(1, workingChunks.count)<br> )<br> GRLog.p(&quot;ResearchAssistant&quot;, &quot;[chat] concatenated summaries into single entry; retrying…&quot;)<br> attempt += 1<br> continue<br> } else {<br> throw error<br> }<br> }<br> }<br>}<br>// ResearchSummaryService.swift (helpers used by both exec summary + chat)<br>func clampAggregate(_ chunks: [SummaryChunkPayload], maxCharacters: Int) -&gt; [SummaryChunkPayload] {<br> chunks.map { chunk in<br> var c = chunk<br> if c.summary.count &gt; maxCharacters {<br> c.summary = String(c.summary.prefix(maxCharacters))<br> }<br> return c<br> }<br>}<br>func concatenateAggregate(_ chunks: [SummaryChunkPayload], maxCharacters: Int) -&gt; [SummaryChunkPayload] {<br> let joined = chunks<br> .map { $0.summary }<br> .joined(separator: &quot; &quot;)<br> let trimmed = String(joined.prefix(maxCharacters))<br> return [SummaryChunkPayload(chunk: 0,<br> offsets: OffsetRange(start: 0, end: trimmed.count),<br> summary: trimmed)]<br>}</pre><h3>Why it matters</h3><p>For us, reading in augmented reality isn’t about speed or spectacle. It’s about space — about giving ideas room to breathe.</p><blockquote>Gravitas Reader is an experiment in that direction. It’s a small attempt to make research and reflection feel a bit more physical, and maybe a little more human.</blockquote><blockquote><em>Try it out for on Apple TestFlight &gt; </em><a href="https://testflight.apple.com/join/cevgSp8u"><em>Here</em></a></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/230/0*53UAq91bKyNIz8S-.png" /><figcaption>Source: <a href="https://testflight.apple.com/join/cevgSp8u">Apple Test Flight</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9ab6318f4bbb" width="1" height="1" alt=""><hr><p><a href="https://medium.com/antaeus-ar/gravitas-reader-turns-ar-into-a-multi-room-content-magazine-now-lets-look-under-the-hood-9ab6318f4bbb">Gravitas Reader turns AR into a multi-room content magazine — now let’s look under the hood</a> was originally published in <a href="https://medium.com/antaeus-ar">Antaeus AR</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building Gravitas Dark Matter: A Retro-Futurist Shooter for Apple Vision Pro]]></title>
            <link>https://medium.com/@richfallatjrshop/building-gravitas-dark-matter-a-retro-futurist-shooter-for-apple-vision-pro-7389c6fcd6a0?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/7389c6fcd6a0</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[pipeline]]></category>
            <category><![CDATA[apple-vision-pro]]></category>
            <category><![CDATA[case-study]]></category>
            <category><![CDATA[game-development]]></category>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Thu, 28 Aug 2025 07:59:33 GMT</pubDate>
            <atom:updated>2025-08-28T08:28:45.559Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Apple Vision Pro hasn’t yet found its “Pong moment.” I set out to build it myself — a gaze-driven space shooter that feels part retro codec screen, part cosmic simulation.</em></p><iframe src="https://cdn.embedly.com/widgets/media.html?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DJ07gGowsapg&amp;type=text%2Fhtml&amp;schema=youtube&amp;display_name=YouTube&amp;src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FJ07gGowsapg%3Ffeature%3Doembed" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/6143575a102ffccc1db22c26286fe63e/href">https://medium.com/media/6143575a102ffccc1db22c26286fe63e/href</a></iframe><h3>Introduction</h3><p>I’ve been a huge fan of 80.lv since 2017 — the site shaped much of my approach to technical art, pipelines, and art direction. After years of working in look development at Disney Animation, Sony Pictures Imageworks, and running my own experiments, I wanted to share my latest finished project: <strong>Gravitas Dark Matter</strong>, a gaze-driven space shooter built natively for Apple Vision Pro.</p><p>This is both a design diary and a statement: you don’t need massive assets or teams to create something that feels cinematic. You can lean on <strong>procedural systems</strong> and a strong <strong>aesthetic direction</strong> to get there.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cE1T0lQZrDB56Ok78AO-MQ.jpeg" /><figcaption>Logo treatment and early title screen for Gravitas Dark Matter.</figcaption></figure><h3>The Premise</h3><p>The player commands a drifting ship across orbital fields, defending against collapsing stars and supernova mines. The entire experience is driven by <strong>eye gaze + pinch targeting</strong> and a <strong>custom gravitational solver (Gravitas Engine)</strong>.</p><p>The mood nods to <em>Metal Gear Solid’s</em> codec UI, reinforced by synth-wave sound. My goal was to make the experience both cinematic and minimal — no heavy assets, just math and motion.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*JBA079i3xayO0r6z2Pg4wQ.gif" /><figcaption>Gameplay supports full 6DoF movement, letting players drift naturally in space.</figcaption></figure><h3>The Problem</h3><p>Most XR titles lean heavily on prebuilt assets. On Vision Pro, that slows things down. I wanted to see if I could:</p><ul><li>Deliver <strong>cinematic spectacle</strong> with almost no preloaded assets.</li><li>Keep performance high on Vision Pro.</li><li>Anchor the player with a readable, nostalgic HUD that works in XR.</li></ul><h3>The Technical Approach</h3><ol><li><strong>Procedural Gameplay</strong><br> Every orbit and explosion is solved in real-time. Bodies interact using Gravitas Engine’s gravitational averaging, giving natural-feeling cosmic motion.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tAx_uk0ydAyOZBxozV7jOw.png" /><figcaption>The Apple Immersive Environment boosts lighting realism and cosmic scale.</figcaption></figure><p><strong>2. Minimal Assets</strong><br> All FX — bursts, trails, orbits — are procedural. The package size stays tiny, but the motion looks alive.</p><p><strong>3. Codec HUD</strong><br> A head-anchored HUD inspired by MGS codec calls. Clean lines, readable in motion, retro feel.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*WsOGDlLh3RAiKHKNyzHjaQ.gif" /><figcaption>Codec-style HUD design, head-anchored for stability and virtual haptic feedback.</figcaption></figure><p><strong>4. Built for visionOS</strong></p><ul><li>Swift + RealityKit core.</li><li>60 handcrafted levels.</li><li>Multiverse mode with deterministic seeds.</li><li>Seamless mixed reality and full immersion.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*K4vGyvORoRP5NSYAG5l-DQ.gif" /><figcaption>Mixed Reality mode adapts to room-scale play, blending virtual nodes with real walls.</figcaption></figure><h3>Results</h3><p>The game runs stably on Vision Pro and looks larger-than-life despite the minimal footprint. Watching a supernova mine collapse or an orbital swarm whip around is both physical and readable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Q7iOTduMaI8KMdTK59AQkg.jpeg" /><figcaption>Testing Gravitas Dark Matter inside Apple’s Moon Immersive Environment.</figcaption></figure><h3>Lessons Learned</h3><ul><li>Procedural &gt; asset-heavy pipelines in XR.</li><li>Eye gaze + pinch feels intuitive when tuned well.</li><li>Nostalgic aesthetics can make futuristic devices more approachable.</li></ul><h3>What’s Next</h3><p>I want to push the synth-wave vibe further with sound design, refine multiverse replay, and explore public release.</p><h3>Closing</h3><p><em>Gravitas Dark Matter</em> started as a technical test. It became proof that <strong>cinematic XR experiences don’t require massive pipelines</strong>. With the right math and mood, even a solo developer can create something that feels alive.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FS2TTMp9ftbU%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DS2TTMp9ftbU&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FS2TTMp9ftbU%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/3833e3cad08022c83585062750961bfa/href">https://medium.com/media/3833e3cad08022c83585062750961bfa/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7389c6fcd6a0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gravitas Dark Matter: designing new interactions for spatial computing on Apple Vision Pro]]></title>
            <link>https://medium.com/design-bootcamp/gravitas-dark-matter-redefining-spatial-play-on-apple-vision-pro-962c5b449935?source=rss-e21153971cf3------2</link>
            <guid isPermaLink="false">https://medium.com/p/962c5b449935</guid>
            <category><![CDATA[spatial-computing]]></category>
            <category><![CDATA[apple-vision-pro]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[game-design]]></category>
            <dc:creator><![CDATA[Rich Fallat]]></dc:creator>
            <pubDate>Tue, 26 Aug 2025 08:15:46 GMT</pubDate>
            <atom:updated>2025-08-26T22:11:57.021Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A case study on Eye-Pinch Targeting and stylization in Apple Vision Pro.</em></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FS2TTMp9ftbU%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DS2TTMp9ftbU&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FS2TTMp9ftbU%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/01ef3841a0f3459b975754456562d60b/href">https://medium.com/media/01ef3841a0f3459b975754456562d60b/href</a></iframe><h3>Introduction</h3><p>Most early Vision Pro apps mimic the real world: photoreal windows in virtual space, menus that look like physical objects. At Disney Animation, Sony Pictures Imageworks, and beyond, I’ve learned that <strong>design isn’t about copying reality — it’s about creating intentional new ones.</strong></p><p>That mindset led me to build <strong>Gravitas Dark Matter</strong>, a stylized, physics-driven arcade experience for Apple Vision Pro. On the surface it’s a game. But underneath, it’s an exploration of new input models, a critique of photoreal mimicry, and a study in how <strong>HUDs can become characters instead of decoration.</strong></p><p>Apple previewed many of these interaction patterns during WWDC 2025 when it announced visionOS 26 <a href="https://www.apple.com/newsroom/2025/06/visionos-26-introduces-powerful-new-spatial-experiences-for-apple-vision-pro/">Apple Newsroom</a>, underscoring how quickly the design space is evolving.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4oiXa9IhT31c5X3GjHOnzw.jpeg" /><figcaption>Mixed Immersion Room Mode shown in the player’s environment.</figcaption></figure><h3>Eye-Pinch Targeting: A New Input Model</h3><ul><li>Eyes act as the crosshair.</li><li>A pinch gesture confirms intent.</li><li>The head provides gross aim, while the eyes deliver micro-accuracy.</li></ul><p>This mechanic feels inevitable once you try it. But it surfaced new design challenges: <strong>eye strain, off-center fatigue, and precision tuning.</strong></p><ul><li>Solution: adaptive dampening and a HUD reticle that anchors feedback.</li><li>UX takeaway: <strong>when gaze is the cursor, comfort and recovery cues matter as much as accuracy.</strong></li></ul><p>Similar challenges have been documented in research on gaze-and-pinch interaction in XR, which highlight how fragile comfort and timing can be in gaze-driven input <a href="https://arxiv.org/abs/2401.10948">arXiv:2401.10948</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*K4vGyvORoRP5NSYAG5l-DQ.gif" /><figcaption>HUD shape language designed to guide the player’s gaze to the center of the frame.</figcaption></figure><h3>Stylization vs Photorealism in Vision Pro</h3><p>Photoreal environments feel familiar, but they can also limit imagination. Stylization demands more upfront design — choosing shapes, contrasts, and abstractions deliberately.</p><p>In Gravitas Dark Matter, neon arcs and abstract cosmology create <strong>clarity in chaos</strong>. Stylization isn’t easier than realism — it’s harder, because every visual choice must serve cognition and feedback.</p><p>UX takeaway: <strong>stylization in XR isn’t an aesthetic shortcut; it’s a tool for focus, legibility, and player emotion.</strong></p><p>This tension mirrors long-standing UX debates about abstraction and realism in interface design <a href="https://uxplanet.org/designing-for-the-immersive-world-a-ux-designers-guide-to-ar-vr-and-xr-c2414802be59">UX Planet</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*7Ex2y46DvgD57sj3C1BAcg.gif" /><figcaption>Eye-Pinch Targeting mechanic: gaze as crosshair, pinch to fire.</figcaption></figure><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FDVZ7D_hve9Y%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DDVZ7D_hve9Y&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FDVZ7D_hve9Y%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/f127b75b3a59d53d9e0d0222b8def149/href">https://medium.com/media/f127b75b3a59d53d9e0d0222b8def149/href</a></iframe><h3>Emergent Physics as Design System</h3><p>Rather than scripting difficulty, Gravitas runs on a custom gravitational solver:</p><ul><li>Black holes multiply.</li><li>Wormholes scatter bodies unpredictably</li><li>Supernova mines chain-react.</li></ul><p>This creates a difficulty curve designed by math, not by scripted behavior. No two playthroughs feel alike.</p><p><strong>Physics is the interface.</strong> For example, one of the most thrilling mechanics is the race against time as black holes absorb rogue bodies. As a body nears the event horizon, its mass is reduced while its velocity and acceleration spike. The result is a slingshot effect that abides by the laws of physics.</p><p>It’s not guaranteed — sometimes the body vanishes, sometimes it flings out at hundreds of miles an hour. But in that chaos lies the game: the player has the chance to snipe the body midair and save it. Watching a world naturally slingshot past a black hole, only to be rescued at the last second, transforms physics from background simulation into the <strong>core drama of play.</strong></p><p><strong>UX takeaway:</strong> sometimes the best level designer is the system itself. Emergent behaviors challenge both player and designer expectations, creating moments of surprise, tension, and agency.</p><p>This idea has precedent in game design, where emergent systems often create more engaging drama than scripted mechanics <a href="https://www.gamedeveloper.com/design/examining-emergent-gameplay">Game Developer</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*JBA079i3xayO0r6z2Pg4wQ.gif" /><figcaption>Custom physics solver with speculative wormhole effects.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sET8mtjwV8swGwOKZ2XDxQ.jpeg" /><figcaption>Black holes attracting bodies with Newtonian formulas, causing neon explosions.</figcaption></figure><h3>Modes of Play</h3><p><strong>Room Mode</strong><br> Designed for a standard living space — ~10 × 8 × 10 ft, with an 8 ft ceiling. This makes Gravitas approachable in almost any home setup.</p><p><strong>Arena Mode</strong><br> Expands the playfield to 30 × 30 × 100 ft outdoors. For the first time, Gravitas turns spatial computing into a <em>sport</em>, with players moving like athletes through a field-sized volume.</p><p><strong>Galaxy Full Immersion Mode</strong><br> The widest arena yet. Here the player floats at the center of a physical starfield rendered with true parallax. Hazards spawn not just around you but above and beneath you, creating full 360° play. The sensation is less like standing in a room and more like being suspended in outer space.</p><p><strong>Arcade Mode</strong><br> A classic run of 60 handcrafted levels tuned for about 15 minutes of play. Focused, rhythmic, and replayable.</p><p><strong>Multiverse Mode</strong><br> Endless galaxies seeded with shifting constants — gravity, acceleration, density — producing universes that never repeat.</p><ul><li>Each galaxy spans 10 levels, gated by the length of a track.</li><li>The <strong>Neon Orchestra</strong> glows with the music, mines swelling with bass and stars flaring with treble.</li><li>250+ levels already composed, with infinite potential extension.</li></ul><p><strong>UX takeaway:</strong> Each mode is a different lens on spatial play:</p><ul><li><strong>Room</strong> = accessible</li><li><strong>Arena</strong> = athletic</li><li><strong>Galaxy</strong> = cosmic embodiment</li><li><strong>Arcade</strong> = rhythmic mastery</li><li><strong>Multiverse</strong> = infinite experimentation</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tAx_uk0ydAyOZBxozV7jOw.png" /><figcaption>All Apple immersive backgrounds available in visionOS 26.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MAwTRofWWJOlsPBse5jonA.jpeg" /><figcaption>End screen easter egg highlighting the Laws of Gravity.</figcaption></figure><h3>HUD With Visual Haptics</h3><p>The cockpit-style HUD doesn’t just display information. It reacts. Impacts scramble it, shake it, and simulate tactile feedback — <strong>without extra hardware.</strong></p><p>Interface as feedback loop. Interface as character.</p><p>UX takeaway: <strong>in spatial computing, UI can be more than information — it can embody the sensation of impact.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*WsOGDlLh3RAiKHKNyzHjaQ.gif" /><figcaption>HUD visual haptics and player feedback animations in action.</figcaption></figure><h3><strong>HUD Shape Language</strong></h3><p>Unlike HUDs in sci-fi movies — often shown as flat plates of glass — we quickly realized the power of a <strong>volumetric HUD.</strong></p><p>In <em>Gravitas Dark Matter</em>, the HUD isn’t just a transparent overlay. Its geometry has depth, with arcs and glyphs designed to <strong>lead the eye back to center frame.</strong> This complements the core mechanic (your eyes as the crosshair), ensuring that in the middle of cosmic chaos, your attention is anchored.</p><p>Where “visual haptics” make the HUD feel tactile, <strong>shape language makes it directive</strong>: the interface itself subtly manages perception.</p><p>UX takeaway: <strong>HUD design in spatial computing isn’t only about legibility — it can guide attention, balance motion, and literally train where the player looks.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0M39CU8Riq98A3VSSsHG7g.jpeg" /><figcaption>Full Immersion Galaxy Mode with Vision Pro hardware and product branding.</figcaption></figure><h3>Why This Matters</h3><p><strong>Gravitas Dark Matter</strong> is more than a game — it’s an experiment in spatial design:</p><ul><li><strong>Input</strong> → what if your eyes + pinch became the controller?</li><li><strong>Visuals</strong> → what if stylization offered a designed alternative to photoreal mimicry?</li><li><strong>Physics</strong> → what if difficulty emerged from math itself, not scripted AI?</li></ul><p>The Vision Pro ecosystem is just beginning. Gravitas is proof that spatial computing doesn’t have to imitate what came before — it can invent entirely new ways to play.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=962c5b449935" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/gravitas-dark-matter-redefining-spatial-play-on-apple-vision-pro-962c5b449935">Gravitas Dark Matter: designing new interactions for spatial computing on Apple Vision Pro</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>