<?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 Shrivasporas on Medium]]></title>
        <description><![CDATA[Stories by Shrivasporas on Medium]]></description>
        <link>https://medium.com/@shrivasporas?source=rss-84d4ddb4d14f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Stories by Shrivasporas on Medium</title>
            <link>https://medium.com/@shrivasporas?source=rss-84d4ddb4d14f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 17:58:33 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@shrivasporas/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[I built a free way to send large files. Peer-to-peer broke on mobile data. Mine works flawlessly :)]]></title>
            <link>https://medium.com/@shrivasporas/i-built-a-free-way-to-send-large-files-peer-to-peer-broke-on-mobile-data-mine-works-flawlessly-a88b5761064e?source=rss-84d4ddb4d14f------2</link>
            <guid isPermaLink="false">https://medium.com/p/a88b5761064e</guid>
            <category><![CDATA[file-sharing]]></category>
            <category><![CDATA[webrtc]]></category>
            <category><![CDATA[build-in-public]]></category>
            <category><![CDATA[indie-hacking]]></category>
            <category><![CDATA[programming]]></category>
            <dc:creator><![CDATA[Shrivasporas]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 17:21:45 GMT</pubDate>
            <atom:updated>2026-11-18T17:21:45.507Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Direct transfers fail on mobile networks, and a relay costs money. How </em><a href="https://boltshare.in/"><em>BoltShare </em></a><em>got around both, and how people found it.</em></p><p>Sending a large video to friends or family used to go like this for me. Everyone has a different phone, some on Android and some on iPhone, so AirDrop and Quick Share don’t work between all of them. Some people don’t want the video on WhatsApp or Telegram because of the compression, and some files are too large to send there anyway. The usual next option is Google Drive, where I upload the file and create a link. A free account stops at 15 GB, and most people have already filled theirs. The techy ones use Mega or another cloud upload, but those have limits too, and the file still has to be uploaded in full before anyone can start downloading it.</p><p>Then there are the web tools that use peer-to-peer, where the file goes straight from one device to the other. They work well when both devices are on the same network. Send to someone on a mobile network and forget it, it takes forever.</p><p>So in February 2026 I started building my own, mostly to look deep into why peer-to-peer wasn’t working the way I wanted. It is called <a href="https://www.boltshare.in/">BoltShare</a>, and it is a free way to send large files from one browser to another, with no login, no app and no size limit.</p><p>I used AI to help me write this article, working from my own notes and the posts on my site.</p><h3>Why peer-to-peer fails on mobile data</h3><p>The first commit is dated February 3, and its message is one word: stable. It wasn’t, and two days later I was already fixing how the browser stored incoming files. The plan was simple, though. Two browsers connect straight to each other with WebRTC, and the file goes across with no server in the middle. Nothing is uploaded first, so there is nothing to wait for and no reason to cap the size.</p><p>Three weeks in, my own tool had the problem I had seen in the others. Some pairs of devices couldn’t reach each other at all, and I had to add a TURN server, which is the fallback WebRTC has for this.</p><p>I went deep into networking to find out what was wrong with mobile networks. A direct connection needs each browser to work out how it looks from the outside, and then reach the other one through both routers. On home wifi this usually works. On mobile data, thousands of phones share one public address (it is called carrier-grade NAT), and the trick that direct connections rely on fails there. A phone that moves from wifi to mobile data also changes its address in the middle of a transfer, so the route has to be worked out again. It is a limit of the technology, and I can’t do anything about it from my side.</p><h3>A relay that a free tool can afford</h3><p>The only other option is a relay, a server in the middle that passes the file along. But a free tool that doesn’t charge its users can’t survive the cost of a relay.</p><p>So I tried a different approach and used Cloudflare’s Durable Objects. Both browsers open an ordinary secure WebSocket to a small object on Cloudflare’s network, one per transfer. It forwards each piece of the file the moment it arrives and keeps nothing. It is the same kind of connection every website uses, so the networks that block peer-to-peer let it through.</p><p>The first version went in on July 1 and was 177 lines of code. It needed a backpressure fix and a race condition fix before the day was over, and two commits from that day have a full stop as their entire message.</p><p>After that I engineered the whole flow around it. The file goes across in 1 MiB pieces, and drops to 256 KiB pieces if the network rejects the bigger ones. The receiver confirms every 4 MB it has written, and the sender is never allowed to get more than 24 MB ahead of what has been confirmed. If a connection drops, both sides reconnect and carry on from the last byte the receiver confirmed, so a transfer doesn’t start over. It is now called Bolt Tunnel and it is the default mode.</p><p>Slow networks kept correcting me. Cloudflare bills the relay by the number of messages it handles, so in October I tried to cut the bill by sending bigger pieces, 8 MiB each. On slow connections, one piece took longer to arrive than the timer the receiver used to decide the connection was dead. The receiver closed the socket, reconnected and asked for the same piece again, forever. The next day I went back to 1 MiB.</p><p>Then in early November I ran a speed test on my own PC, with the sender and the receiver in two browser tabs. Direct P2P moved 256 MB at about 14.4 MiB per second and Bolt Tunnel managed about 2.9, which looks like a five to one win for peer-to-peer. With both tabs on one computer, though, the direct data never leaves the machine, while the tunnel’s data goes out to Cloudflare and back. I had measured my own broadband plan.</p><p>I kept Direct P2P as the second mode, because it does two things the tunnel can’t. It is end-to-end encrypted, so no server can read the file, and it is faster when both devices are on the same wifi. The tunnel is encrypted with TLS to and from the relay, and the relay never writes the file to disk.</p><h3>Nobody could find it</h3><p>BoltShare worked without charging anyone money, and I found it faster and more reliable than the tools Google showed me on the first page. Nobody used it, because how would anyone find it on Google?</p><p>I knew this problem already. In college I published a few Android apps on the Play Store. One was HealthMag, something like Flipboard for health, and I had no idea how to get it to people. I was naive about the indie developer thing back then, and BoltShare was heading the same way.</p><p>So this time I worked on being found. I started in June, a month before the relay. The site was a single React page and search engines had very little to read, so I moved it to Astro, where every page is real HTML. I researched what people type into Google, looking for searches where Google doesn’t have much to show them. I found a few and wrote blog posts on BoltShare around them. I added pages that compare BoltShare with its competitors. And I did the usual SEO best practices: internal links between pages, a robots file, a sitemap, an About page and a privacy policy.</p><p>Then I forgot I built it. Four or five months later, people had started using it. It began small, with 7 to 10 people a day. Over about four months that went up to 80 to 100 a day. I added PostHog to see what they were doing. BoltShare asks for a rating after a transfer, and I almost always get 5 stars.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/478/1*y0gSfVIbQaJxNpSFb-wd2g.png" /></figure><p>On November 13, Google sent me this. Search Console counts 89 clicks from Google in June and 908 in October. In the last 28 days, about four in five clicks came from India.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rj54PxE0-jbdayrfrkp_Jw.png" /></figure><h3>How to send large files with it today</h3><p>Open <a href="https://www.boltshare.in/">boltshare.in</a> and drop in a file or a folder. It gives a link and a QR code. The other person opens the link in any browser and the download starts, with no account or app on either side. A 100 GB video folder works the same way a 1 MB PDF does, and the file arrives as the original, without compression. For a device that can’t scan a QR code, like a TV, there is a 6-digit code to type in.</p><p>It has one limit I’d rather say up front. BoltShare never stores the file, so the link only works while the sender’s tab is open. If the other person will download tomorrow, an upload service like WeTransfer or Google Drive is the better choice.</p><h3>Where I am now</h3><p>BoltShare has more than 2000 monthly active users. It is free and supported by ads, which pay for the servers and the relay.</p><p>I know file sharing already existed before I built this. I wanted mine to be reliable even on restricted and slow networks, and that is the part I keep working on. In future I want to add a feature I can charge money for, for the convenience, and I’d like to hear from the people who would use it. If you send large files often, what is the one thing you would pay for? Tell me in the responses.</p><p>Developing was the first step and distribution is the next one. I’m learning that by doing, and this article is part of it.</p><p>BoltShare is at <a href="https://www.boltshare.in/">boltshare.in</a>. If something breaks for you, tell me. I read every message.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a88b5761064e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I’m not a designer. My friend and I built an AI that designs apps into Figma anyway.]]></title>
            <link>https://medium.com/@shrivasporas/im-not-a-designer-my-friend-and-i-built-an-ai-that-designs-apps-into-figma-anyway-a944cf81eddb?source=rss-84d4ddb4d14f------2</link>
            <guid isPermaLink="false">https://medium.com/p/a944cf81eddb</guid>
            <category><![CDATA[ui]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[developer-productivity]]></category>
            <category><![CDATA[figma]]></category>
            <category><![CDATA[indiehackers]]></category>
            <dc:creator><![CDATA[Shrivasporas]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 11:43:47 GMT</pubDate>
            <atom:updated>2026-11-18T11:43:47.500Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Why most AI-made apps look the same in 2026, and what we did about it with Outclass</em></p><figure><img alt="Outclass: design beautiful mobile apps, just by chatting" src="https://cdn-images-1.medium.com/max/1024/1*ON-t4lxUGQ6Eb8yw0sfI8A.png" /></figure><p>I’m a backend developer, and I am not a designer. I do have an eye for design details when I sit down with a screen, though, and that is how a friend and I ended up building Outclass. You describe your app in a sentence or two, it designs the screens, and you copy them into Figma as real layers you can keep working on.</p><h3><strong>Every AI app looks the same</strong></h3><p>If you’ve opened a few AI-made apps this year, you already know the look: a purple gradient, frosted glass cards, the Inter font, and a big centered headline over three boxes. It looks finished at first glance and generic at the second. You can’t show it to a client, and you can’t build on it.</p><p>Part of the purple goes back to Tailwind CSS. Adam Wathan, who created it, half-jokingly apologised in 2025 for making the buttons in Tailwind UI indigo years ago. So much of the web copied those examples, and the models learned from that web.</p><p>A model gives you the most likely answer. Type “a modern, clean app” and you get the average of every app it has seen, because nothing in that prompt says what you want instead.</p><h3><strong>Three kinds of people</strong></h3><p>I’ve written before about three kinds of people in the AI era, and the same three show up in design.</p><p>Some designers are waiting for the bubble to burst. They see the purple gradients and decide AI design will never be good enough.</p><p>Other people have lost hope. They watch an agent build a full screen in seconds and decide there is nothing left for a designer to do.</p><p>I’m in the middle. AI can really do wonders here, but it takes a person to do it. The tools can draw a screen in seconds. Someone still has to decide what that screen should look like and why.</p><h3><strong>What the big tools are doing</strong></h3><p>Figma is moving fast. Figma Make turns a prompt into a working prototype, and most reviews call it good for a rough first version. Figma has also opened its canvas to AI agents: through its MCP server, an agent like Claude Code can write real frames, components and auto layout straight into a Figma file. That feature is free while it’s in beta and will be charged by usage later.</p><p>Most of Figma’s MCP work is built to give coding agents your existing designs as context, so the code they write matches the design. We built Outclass for the other direction, starting from an idea and ending with a design you can keep working on.</p><h3><strong>What we put into Outclass</strong></h3><p>I’ve always been the person who does the tedious part other people skip. As a kid that meant juggling cracks and torrents and tuning the computer until games ran at a good FPS. For Outclass it meant sitting down and learning what good design actually means, then writing it down as rules so the next person doesn’t have to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YecRbyw2m4iPH4fCmyvmHQ.png" /></figure><p>The first rule is that there is no template. A meditation app and a budgeting app shouldn’t look alike, so Outclass first works out what kind of product you’re describing and which screens it really needs. A food delivery app needs a menu and a basket, and a fitness app needs a way to log a workout. Then it picks a visual direction that suits that kind of product, anywhere from calm and editorial to bold and graphic, and it makes sure your next app doesn’t look like your last one.</p><p>The rest are the rules a good designer follows without thinking:</p><blockquote>- Real content instead of “Lorem ipsum”, so you judge a design by what it will actually say.</blockquote><blockquote>- Illustrations drawn for your subject instead of gray placeholder boxes.</blockquote><blockquote>- One type scale, one palette and one set of components across every screen.</blockquote><blockquote>- Navigation only where it belongs, and one clear focal point per screen.</blockquote><p>Once those decisions are fixed, the model has far less room to fall back to the average. Ask it to choose colours and you get the average. Give it the colours and it uses them.</p><p>A first draft is only useful if you can change it, so you can ask for changes in plain words, like “make the cards calmer”, or click any element and set its exact size, spacing, type and colour yourself. Everything snaps to a 4px grid. You can also chat with one element: ask for a change to one card, and the rest of the screen stays as it was.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*LVQiUhn58mvx219KCc2uwg.jpeg" /></figure><h3><strong>Getting it into Figma</strong></h3><p>A lot of AI design tools hand you a picture, or layers so messy that someone has to rebuild them before anyone can use them. When you press Copy to Figma in Outclass and paste, the screen arrives with auto layout, real editable text and vector icons. The layers have names like Header, Card and Tab bar, so nobody has to dig through Frame 12. A designer can polish it from there and a developer can inspect the spacing and colours. You don’t need a Figma plugin, and copying never uses credits.</p><p>We also built an Outclass MCP server. If you already work in Claude Code, Cursor or ChatGPT, you can ask your agent to design a screen, and it sends that screen to Figma the same way.</p><h3><strong>Where I’m at</strong></h3><p>AI filled in the broken parts of my knowledge. I never went to design school, and now my friend and I have a product built to give people a first draft that a designer would be glad to keep working on. We got there by doing the slow part once and putting it into a tool.</p><p>I think AI will take curious people to real heights and crush the people who just finish off their to-do list. In design, the to-do list version is typing “make an app”, shipping whatever comes back, and wondering why it looks like everyone else’s.</p><p>Right now Outclass designs mobile app screens, and you can try it free at <a href="https://outclass.design/">outclass.design</a>. Landing pages are coming very soon.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a944cf81eddb" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>