<?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 Purna Srivatsa on Medium]]></title>
        <description><![CDATA[Stories by Purna Srivatsa on Medium]]></description>
        <link>https://medium.com/@purnasrivatsa96?source=rss-1e279dc88191------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*f_xwAWB97-z54wCHS4XD7Q.jpeg</url>
            <title>Stories by Purna Srivatsa on Medium</title>
            <link>https://medium.com/@purnasrivatsa96?source=rss-1e279dc88191------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 23:45:39 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@purnasrivatsa96/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[Explore Research Like Following a Treasure Map]]></title>
            <link>https://medium.com/@purnasrivatsa96/explore-research-like-following-a-treasure-map-ed909b46bf8e?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/ed909b46bf8e</guid>
            <category><![CDATA[citations]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[academia]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Fri, 05 Dec 2025 18:20:48 GMT</pubDate>
            <atom:updated>2025-12-05T18:20:48.653Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Stop drowning in browser tabs. Chat with entire citation networks, get answers with exact page numbers, and explore research the way your brain naturally works.</em></p><p>You’re reading a fascinating paper on transformer architectures. It references a foundational 2014 paper on attention mechanisms that sounds important. You open it in a new tab.</p><p>That paper cites three more papers you should probably read. More tabs. Meanwhile, you wonder: “Did anyone improve on this approach? What criticisms emerged? How is this being applied today?”</p><p>You click through to papers citing the original work. 200+ results. You start reading titles, opening more tabs. Twenty minutes later, you have 30 browser tabs open and you’ve lost track of why you opened half of them.</p><p>Reading research shouldn’t feel like falling down a rabbit hole.</p><h3>The Problem: Lost in Citation Limbo</h3><p>The connections between papers — the intellectual lineage of ideas — matters as much as the papers themselves. But traditional tools make you work against your natural curiosity.</p><p>Want to understand how a methodology evolved? You manually trace citations forward through time, paper by paper. Want to know what criticisms emerged? You scan abstracts of citing papers hoping the title gives it away. Want to find applications in your specific domain? Good luck — you’re clicking through hundreds of results.</p><p>Lists can’t answer these questions. Neither can citation graphs alone. You need conversation.</p><h3>Enter Citation Network Chat</h3><p>Today, we’re launching <strong>Citation Network Chat</strong> on Base. It’s a new way to explore research that mirrors how researchers actually think — by asking questions, following threads, and building understanding through conversation.</p><p>Here’s what makes it different.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qfnc7-wgkweAcEB3e97ASA.png" /><figcaption><em>Ask questions about an entire citation network — not just one paper</em></figcaption></figure><h3>Chat with 200+ Papers at Once</h3><p>When you open a paper’s citation network visualization and click “Chat,” something powerful happens behind the scenes. Base loads the complete intellectual context:</p><ul><li>The paper you’re viewing (the focus paper)</li><li>All papers it references (usually 10–50 papers)</li><li>The top 200 papers that cite it (ranked by impact)</li></ul><p>That’s often <strong>200+ papers in a single conversation</strong> — ready to answer your questions. Not just the one paper you’re reading. The entire network of connected research.</p><h3>Ask real questions</h3><p>Instead of clicking through lists, you can ask:</p><ul><li>“How do later papers apply this methodology?”</li><li>“What criticisms have been raised about these results?”</li><li>“Which citing papers focus on clinical applications?”</li><li>“Did anyone improve the computational efficiency?”</li><li>“What foundational work does this build on?”</li></ul><p>The AI reads across all papers in the network and synthesizes answers with precise citations.</p><h3>Citations with Exact Page Numbers</h3><p>Here’s where it gets really useful. Every claim comes with a citation that includes the specific page number.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ozwb3lmah1WpPLStdHa5Cw.png" /><figcaption><em>Citations include exact page numbers — click to jump directly to the source</em></figcaption></figure><p>When Base cites a paper, you see something like this: [2:7]</p><p>That means: <strong>Source 2, page 7.</strong></p><p>Click it, and you jump directly to page 7 of that paper’s PDF. No searching, no scrolling, no guessing. You go straight to where the claim is supported.</p><p>This is how research should work. Make a claim, back it up, show exactly where the evidence is.</p><h3>Add Papers to Your Visualization</h3><p>Sometimes the AI references a paper that’s not currently visible in your citation network. Maybe it’s ranked #47 by citations, so it wasn’t in the initial 30 papers displayed.</p><p>No problem.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rtyFGMPpWxX45_ZOi8aI6g.png" /><figcaption><em>One click adds any cited paper to your visualization</em></figcaption></figure><p>Click “Add to Visualization” and the paper appears instantly in your network, positioned correctly by publication date and topic. Now you can see how it connects to other papers in the citation graph.</p><p>Switch seamlessly between conversation and visualization. Ask questions, discover relevant papers, add them to your network, explore visually. Repeat.</p><h3>A Real Example</h3><p>Let’s say you’re exploring “Attention Is All You Need” (the paper that introduced Transformers). You open the citation network and start a chat.</p><p><strong>You ask:</strong></p><blockquote><em>“What improvements did later papers make to the attention mechanism?”</em></blockquote><p><strong>Base responds:</strong></p><blockquote><em>Several key improvements emerged in subsequent work. Linformer [2:3] introduced a linear complexity attention mechanism, reducing computational requirements from O(n²) to O(n). Reformer [4:7] proposed locality-sensitive hashing to approximate attention patterns, enabling processing of sequences up to 64K tokens. Performer [7:2] used random feature methods to approximate the attention matrix efficiently…</em></blockquote><p>Each citation is clickable:</p><ul><li>Click [2:3] → Opens Linformer paper to page 3</li><li>Click [4:7] → Opens Reformer paper to page 7</li><li>Click [7:2] → Opens Performer paper to page 2</li></ul><p>Don’t see Linformer in your visualization yet? Click “Add to Visualization” on the paper card, and it appears in the network.</p><p>Now you can see how Linformer connects to other efficiency-focused papers, trace its influence forward, and understand where it fits in the evolution of transformer architectures.</p><h3>Why This Matters</h3><h3>For researchers</h3><ul><li>Discover relevant work you might have missed</li><li>Understand how ideas evolved over time</li><li>Find papers that apply methods to your specific domain</li><li>Get exact page numbers for proper citations in your own work</li><li>Save hours of literature review time</li></ul><h3>For students</h3><ul><li>Grasp how papers build on each other</li><li>Find introductory papers vs. advanced technical papers</li><li>Understand criticisms and limitations of methods</li><li>See the “family tree” of ideas in your field</li><li>Build genuine understanding, not just a reading list</li></ul><h3>For the intellectually curious</h3><ul><li>Ask questions in plain language</li><li>Get answers backed by peer-reviewed sources</li><li>Explore research without getting lost</li><li>Follow your curiosity wherever it leads</li></ul><h3>How It Works Under the Hood</h3><h3>The context window</h3><p>When you open a chat, Base loads up to 210 papers into the AI’s context:</p><ul><li>1 focus paper (the one you’re viewing)</li><li>~10 reference papers (what it cites)</li><li>~200 citing papers (what cites it, ranked by citation count)</li></ul><p>This gives the AI complete context about the paper’s intellectual lineage — what came before and what came after.</p><h3>Citation precision</h3><p>We use a dual-citation system:</p><ul><li>[1] = Paper-level citation (for general claims)</li><li>[1:12] = Paper 1, page 12 (for specific claims)</li></ul><p>This ensures you can verify every factual claim by going directly to the source and not just to the paper, but to the exact page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rPz0EuGPbGyhFgGuEeGpyw.png" /><figcaption><em>Click a citation, land on the exact page that supports the claim</em></figcaption></figure><h3>Your chat history persists</h3><p>When you return to a paper’s citation network, your previous conversations are still there. You can pick up where you left off, continue exploring, or start a new line of inquiry.</p><p>The visualization state is ephemeral (resets on page reload), but your research questions and the AI’s answers are preserved. You can always re-add papers from your chat history with one click.</p><h3>Try It Right Now</h3><p>Citation Network Chat is available now for all Base users. No setup required.</p><ol><li>Search for any paper on Base</li><li>Click “Visualize Citations”</li><li>Click the “Chat” button</li><li>Start asking questions</li></ol><p>That’s it. You’re now exploring research the way your brain naturally works by asking questions, following connections, and building understanding through conversation.</p><h3>Research is a Conversation</h3><p>Every paper you read is part of an ongoing dialogue — between ideas, across time, connecting minds separated by decades and continents.</p><p>Citation Network Chat makes that conversation visible, navigable, and explorable. It transforms research from a solitary activity into what it really is: participation in humanity’s collective knowledge-building project.</p><p>We built this tool because we believe research should be joyful. Following your curiosity shouldn’t mean drowning in browser tabs. Discovering connections between ideas shouldn’t require hours of manual searching.</p><p>Try it today. Ask a question. Follow a thread. Discover something unexpected.</p><p><strong>That’s what research is all about.</strong></p><h3>Start Exploring</h3><p>👉 <a href="https://base.com/visualize"><strong>Try Citation Network Chat Free</strong></a></p><p>Explore 240+ million papers.</p><p><em>Have feedback? We’re listening. What questions are you asking? Which features would make your research easier? Let us know at </em><a href="/service/mailto:feedback@base.com"><em>admin@basedid.com</em></a></p><p><strong>About Base:</strong> Base is an AI research assistant that provides verifiable citations with exact page numbers from 240+ million academic papers. We’re building tools to make research more accessible, more collaborative, and more joyful.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ed909b46bf8e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[See the Complete Picture: Visualize How Research Ideas Flow Through Time]]></title>
            <link>https://medium.com/@purnasrivatsa96/see-the-complete-picture-visualize-how-research-ideas-flow-through-time-55df6acabd15?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/55df6acabd15</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[academia]]></category>
            <category><![CDATA[science]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Sun, 30 Nov 2025 18:27:30 GMT</pubDate>
            <atom:updated>2025-11-30T18:27:30.155Z</atom:updated>
            <content:encoded><![CDATA[<p><em>You found the perfect paper. It has 47 references and 156 citations. Now what?</em></p><p>Most tools give you a list. A long, overwhelming list of titles sorted by date or relevance score. You scroll, you click, you open tabs, you lose track. An hour later, you’re drowning in browser tabs with no clear sense of how these papers actually connect.</p><p>Which references were foundational to the field? Which citing papers took the ideas in new directions? How did this research spread across disciplines? Lists can’t answer these questions.</p><h3>Research is a Conversation Across Time</h3><p>Every paper exists within a web of intellectual dialogue. It builds on what came before and shapes what comes after. Understanding this context transforms how you engage with research.</p><p>When you can see that a 2015 machine learning paper cites foundational statistics work from the 1990s alongside recent computer vision papers, you understand something important about how the field evolved. When you notice that a finance paper is being cited by researchers in public health and education, you’re seeing ideas cross disciplinary boundaries in real time.</p><p>This context matters. It helps you identify seminal works, spot emerging trends, and understand where your own research fits in the larger conversation.</p><h3>Mapping the Flow of Ideas</h3><p>Base’s citation network visualization transforms abstract citation data into a spatial map of research influence. Paper citation network overview showing focus paper in center, references on left (blue lines), citing papers on right (orange lines), organized by topic rows with year headers</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W3_9QyjrFgwyZORUnsUfTA.png" /><figcaption><em>A paper’s citation network: references flow in from the left, citing works extend to the right</em></figcaption></figure><h3>Time flows left to right</h3><p>Papers are arranged chronologically, so you immediately see the temporal structure of a research area. References appear to the left of your focus paper; citing works appear to the right. The visual layout reveals patterns that lists obscure — you can see at a glance whether a paper drew from recent work or reached back to foundational research decades earlier.</p><h3>Topics organized vertically</h3><p>Papers are grouped by their research domain, field, subfield, and topic. A sticky header keeps you oriented as you navigate, always showing where you are in the taxonomy. This organization reveals cross-disciplinary influence — when citation lines cut across topic boundaries, you’re seeing ideas migrate between fields.</p><p>Zoomed view showing topic sidebar with domain/field/subfield/topic hierarchy, citation lines crossing between different topic rows</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*92hpi6puUurRYeEa5PUTvQ.png" /><figcaption><em>Citation lines crossing topic boundaries reveal interdisciplinary influence</em></figcaption></figure><h3>Connections tell the story</h3><p>Blue lines trace backward to references; orange lines trace forward to citing works. Following these connections shows you the intellectual lineage of ideas. Dense clusters of connections highlight the most influential nodes in the network.</p><h3>Navigating Large Networks</h3><p>Research networks can be sprawling. A well-cited paper might connect to hundreds of other works. Several features help you maintain orientation.</p><h3>The minimap</h3><p>A bird’s-eye view of the entire network. A highlighted rectangle shows your current viewport, and you can click anywhere on the minimap to jump to that region. It’s particularly useful for understanding the overall shape of a citation network before diving into details. Screenshot showing minimap in bottom-right corner with viewport rectangle highlighted, main canvas showing portion of larger network</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AsSe3PA9t-K06IsOerpupw.png" /><figcaption><em>The minimap provides overview while you explore details</em></figcaption></figure><h3>The magnifier</h3><p>Solves the zoom paradox. When you’re zoomed out to see the full network, individual paper titles become unreadable. When you’re zoomed in to read details, you lose context. The magnifier gives you both — a floating window that shows a 2x zoomed view of wherever your cursor is positioned. Toggle it on when you need to read titles while maintaining your overview perspective.</p><h3>Dynamic topic labels</h3><p>The sidebar stays synchronized with your view. As you pan and zoom, the hierarchy updates to show the domain, field, and subfield of papers currently visible. You always know where you are in the research landscape.</p><h3>What Patterns Reveal</h3><p>Once you start exploring citation networks visually, certain patterns become immediately apparent.</p><p><strong>Foundational vs. incremental work.</strong> Papers with many outgoing orange lines (heavily cited) that connect to diverse topics are often foundational contributions. Papers that cite many references but have few citations might be comprehensive reviews or newer work that hasn’t yet accumulated citations.</p><p><strong>Disciplinary bridges.</strong> When you see citation lines crossing multiple topic rows, you’re looking at interdisciplinary influence. These bridges often represent innovative applications or theoretical connections between fields.</p><p><strong>Research trajectories.</strong> The temporal layout reveals how quickly ideas spread. Some papers are cited immediately and intensively for a few years. Others accumulate citations slowly over decades. The visual pattern tells you something about the nature of the contribution.</p><p><strong>Clusters and gaps.</strong> Dense regions of the network indicate active research areas. Sparse regions might represent opportunities — questions that haven’t been fully explored or connections that haven’t been made.</p><h3>From Overview to Insight</h3><p>The visualization isn’t meant to replace reading papers — it’s meant to help you decide which papers to read and in what order. Start with the overview, identify the structural patterns, then drill down into the specific papers that occupy key positions in the network.</p><p>A paper that bridges two topic clusters might be worth reading even if its title didn’t initially catch your attention. A heavily-cited older paper that you hadn’t encountered might fill a gap in your understanding. A recent paper positioned at the edge of the network might represent the current frontier of the field.</p><p>The goal is informed navigation. When you can see the shape of a research conversation, you can participate in it more effectively.</p><h3>Try It Yourself</h3><p>Search for any paper in our index of 240+ million works. Click “Visualize Citations” to explore its intellectual context — the works it builds on and the research it inspired.</p><p>See where ideas come from. See where they go.</p><p><a href="https://basedid.com/visualize"><strong>Try Citation Visualization Free →</strong></a></p><p><em>Base is an AI-powered research platform with verifiable, page-level citations. Search 240+ million papers, build persistent repositories, and chat with AI that tells you exactly which page to verify.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=55df6acabd15" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[See Your Research, Don’t Just Search It: How Visualization Transforms Literature Review]]></title>
            <link>https://medium.com/@purnasrivatsa96/see-your-research-dont-just-search-it-how-visualization-transforms-literature-review-463dbed9b5a1?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/463dbed9b5a1</guid>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[academia]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[science]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Thu, 27 Nov 2025 12:45:20 GMT</pubDate>
            <atom:updated>2025-11-27T13:54:04.494Z</atom:updated>
            <content:encoded><![CDATA[<h3>Why staring at a list of papers isn’t working — and what to do instead.</h3><p>Every researcher knows the feeling: you’ve gathered 50 papers on your topic, they’re sitting in a folder somewhere, and you have no idea how they connect. Which paper started this line of inquiry? What are the competing approaches? Where are the gaps?</p><p>Lists don’t answer these questions. Visualization does.</p><h3>The Problem with Traditional Literature Review</h3><p>Most research tools give you one thing: a search bar and a list of results. You type in keywords, get papers ranked by “relevance,” and start the painful process of reading abstracts one by one.</p><p>But research isn’t linear. It’s a web of ideas, citations, methodologies, and disciplines that influence each other across decades. Viewing this as a flat list is like trying to understand a city by reading street names alphabetically.</p><h3>A New Way to See Your Research</h3><p>Base’s Taxonomy-Based Citation Network shows you the shape of your research at a glance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XM5QjctMybiSXcAYW5FiDw.png" /><figcaption>A repository visualized: papers organized by taxonomy hierarchy with details panel</figcaption></figure><p>Here’s what you see when you visualize a repository:</p><p><strong>Taxonomy hierarchy: Domain → Field → Subfield → Topic</strong></p><p>Papers are organized into a tree structure based on their research area. A paper on “Nutrient Profile of Native Woody Species” sits under Life Sciences → Agricultural and Biological Sciences → Agronomy and Crop Science → Ruminant Nutrition and Digestive Physiology. You immediately see where your research lives in the broader landscape — and which disciplines your repository spans.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yOBDsav5C1r2gnWthJRdEw.png" /><figcaption><em>Zoomed in: The taxonomy tree expands from field to subfield to topic, with papers at the leaves</em></figcaption></figure><p><strong>Papers sorted by year within each topic</strong></p><p>Within each topic, papers are arranged chronologically — oldest to newest. This lets you trace how ideas evolved within a specific research area, from foundational work to recent developments.</p><p><strong>Color coding: Taxonomy categories</strong></p><p>Each field gets its own color, making it easy to spot when your research spans multiple disciplines — often where the most interesting insights live.</p><h3>No PhD Required</h3><p>Here’s what makes this different from other research tools: <strong>you don’t need to be an expert to use it.</strong></p><p>The visualization uses simple, universal concepts — folders, trees, color-coded categories. There’s no jargon to learn, no complex query syntax, no discipline-specific knowledge required to get started.</p><p>Curious about climate science but you’re a software engineer? The taxonomy shows you exactly where papers sit: Earth Sciences → Environmental Science → Climate Change → Carbon Sequestration. You instantly understand the landscape without reading a single abstract.</p><p>Wondering how AI and medicine intersect? Watch the visualization light up with papers spanning multiple fields — connections you’d never find through keyword search.</p><p><strong>The structure does the explaining for you.</strong> Domains, fields, subfields, and topics create a universal map that works whether you’re exploring quantum physics or Renaissance art history.</p><p>This matters because the best insights often come from outsiders — people who bring fresh eyes to a field. Traditional research can be overwhelming for newcomers with walls of jargon and assumed knowledge. Base’s visualization welcomes them.</p><h3>Beyond Visualization: The Research Command Center</h3><p>Visualization is powerful, but it’s part of something larger. Base is designed as a persistent home for your research — not a tool you visit once and forget.</p><p><strong>Repository-centric workflow</strong></p><p>Every paper you save lives in a repository organized by your research questions. Come back weeks later and pick up exactly where you left off.</p><p><strong>AI chat with page-level citations</strong></p><p>Ask questions about your research and get answers with exact page numbers — not just paper references. When the AI says “Smith et al. found a 40% reduction,” it tells you <em>which page</em> to verify. No more hunting through 30-page PDFs.</p><p><strong>240+ million papers indexed</strong></p><p>Search across OpenAlex and Semantic Scholar databases. Filter by year, field, topic, or access type.</p><h3>Coming Soon</h3><ul><li><strong>Citation relationship mapping:</strong> Lines between papers showing who cited whom — so you can trace the intellectual lineage of ideas and see which foundational papers sparked entire research directions.</li><li><strong>Multimodal understanding:</strong> Extract information from figures, tables, and diagrams in papers — not just text. Ask about a chart and get an actual answer.</li></ul><h3>Who This Is For</h3><p><strong>Serious researchers</strong> doing literature reviews, mapping dissertation landscapes, or preparing grant applications.</p><p><strong>Curious learners</strong> exploring a new field, following an interest, or trying to understand how disciplines connect.</p><p><strong>Professionals</strong> who need to quickly get up to speed on research outside their expertise — consultants, journalists, policy analysts, product managers.</p><p><strong>Students</strong> at any level who want to see the big picture before diving into the details.</p><p>You don’t need to be an expert to explore like one.</p><h3>Try It Yourself</h3><p>Upload your papers or search for a topic. Build a repository. Then hit “Visualize” and watch your research take shape.</p><p>No setup. No jargon. Just clarity.</p><p><strong>→ Start visualizing your research at </strong><a href="http://basedid.com/visualize"><strong>basedid.com/visualize</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=463dbed9b5a1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[True Understanding Transcends Language]]></title>
            <link>https://medium.com/@purnasrivatsa96/true-understanding-transcends-language-93089ed51ab6?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/93089ed51ab6</guid>
            <category><![CDATA[thinking]]></category>
            <category><![CDATA[learning]]></category>
            <category><![CDATA[language]]></category>
            <category><![CDATA[philosophy]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Mon, 07 Apr 2025 01:01:30 GMT</pubDate>
            <atom:updated>2025-04-07T01:01:30.989Z</atom:updated>
            <content:encoded><![CDATA[<p>Why the deepest insights don’t live in words — and how pseudo-language reveals the shape of thought.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*btMaCGecwr7BKS0k1ano8Q.jpeg" /><figcaption>image from — <a href="https://art.rtistiq.com/en-us/blog/famous-abstract-art-and-artists">https://art.rtistiq.com/en-us/blog/famous-abstract-art-and-artists</a></figcaption></figure><p><em>“This story was written with the assistance of an AI writing program.”</em></p><p><strong>“I understand it, but I can’t explain it.”</strong><br> We’ve all been there. That quiet, frustrating space where clarity lives in your mind but collapses when you try to express it. Maybe it’s a complex algorithm, a philosophical idea, or just a feeling — and no matter how precise your words are, they still feel like shadows of what you truly mean.</p><p>This is not a failure of intelligence. It’s a limitation of language.</p><h3>The Shape of Thought</h3><p>At a certain level, <strong>true understanding transcends language</strong>. When you really grasp an idea — I mean deeply, intuitively — it detaches from the domain of vocabulary and syntax. The idea lives as a kind of <strong>mental object</strong>: flexible, fluid, and convertible into whatever form is needed.</p><p>You can express it in plain speech.<br> You can write it in code.<br> You can draw it, model it, metaphor it.<br> You can reframe it to teach a child or an expert.<br> That’s because the core idea isn’t tied to its original form.</p><p>It’s become <strong>yours</strong>.</p><h3>Enter: Pseudo-Language</h3><p>In that space, you begin speaking in what I call <strong>pseudo-language</strong> — a mental scaffolding where logic and meaning exist without strict grammar. It’s like pseudocode for the mind. A hybrid language where symbols, metaphors, concepts, and logic blend into one.</p><ul><li>You don’t <em>remember</em> the definition. You <em>see</em> the relationships.</li><li>You don’t rely on memorized terms. You rely on cause and effect.</li><li>You don’t need to explain with precision. You just need a bridge to the other person’s understanding.</li></ul><p>And once you’re there, you’re free. You’re no longer a prisoner of domain-specific jargon or linguistic bottlenecks.</p><h3>Beyond Words: Echoes from Philosophy and Science</h3><p>This isn’t just abstract musing — it has deep roots:</p><ul><li><strong>Einstein</strong> believed if you truly understand something, you should be able to explain it simply.</li><li><strong>Wittgenstein</strong> argued that the limits of language are the limits of our world — but hinted that meaning often exists just beyond those limits.</li><li><strong>Zen Buddhism</strong> teaches that direct experience can’t be fully verbalized — that enlightenment lives beyond words.</li><li><strong>Neuroscience</strong> increasingly shows that much of human cognition is non-verbal — visual, spatial, embodied.</li></ul><p>The message is clear: <strong>language is a tool, not a container</strong>. It can point to truth, but it isn’t the truth itself.</p><h3>Why This Matters (Especially Today)</h3><p>In a world dominated by content, code, and constant communication, we’re trained to optimize for <em>how well we say things</em>. But we should spend more time asking:<br> <strong>Do we understand it beyond how we say it?</strong></p><p>True understanding is portable.<br> It transcends disciplines.<br> It can morph across metaphors.<br> It <em>cuts across domains</em>.</p><p>It’s what allows a great teacher to translate quantum physics into cartoons.<br> It’s what lets an artist convey grief without a single word.<br> It’s what enables us to build bridges between radically different minds.</p><h3>Final Thought</h3><p>So the next time you struggle to explain something you <em>know</em> you understand — don’t be discouraged. You might just be closer to mastery than you think.</p><p>Because when ideas become fluid, expressive, and language-agnostic…<br> You’re not just speaking.<br> <strong>You’re thinking in pseudo-language.</strong></p><p>And that’s the language of true understanding.</p><p>Every word above after “<em>“This story was written with the assistance of an AI writing program.” </em>was AI generated. Every word from here is not. And let me explain why i chose to formulate this article in this way.</p><p>I got this thought about the separation of idea/understanding and the language needed to express it. And i tweeted(do we say X’ed now?) about it.</p><iframe src="https://cdn.embedly.com/widgets/media.html?type=text%2Fhtml&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;schema=twitter&amp;url=https%3A//x.com/PurnaSrivatsa/status/1909018364900786681&amp;image=" width="500" height="281" frameborder="0" scrolling="no"><a href="https://medium.com/media/03724759d67d721c069d2ebccc93267b/href">https://medium.com/media/03724759d67d721c069d2ebccc93267b/href</a></iframe><p>I decided to write a medium article expanding the idea and turned to chatGPT to take it’s input and thoughts about the tweet. I wanted chatGPT’s conscious understanding of the extensive and comprehensive historical, philosophical and cognitive science texts to translate the idea into a form that’s easily pluggable into existing world wisdom. The conversation quickly turned into chatGPT proposing to write the article itself. It gave the above article, which is exactly the kind of fleshing out and expansion, i wanted, of the core idea.</p><p>Because the story is about abstract and language agnostic forms of ideas and their importance over articulated versions, i decided to incorporate the message into the method in which the story itself is delivered too.</p><p>ChatGPT’s articulated version is an apt, consumable form of the abstract. My intention is to convey that ideas matter more than words. That mastery of transcoding and registering thoughts in their pure, abstract and language free forms is prerequisite to bi-directionally translate to and from their language versions so that they align and capture the pseudo language form accurately.</p><p>By reaching a state where i can pseudo-language my idea, I could sign off , and convert the language form(that chatGPT provided) of the idea traceably into it’s abstract pseudo-language form.</p><p>I do acknowledge that verbalizing and articulating abstract ideas into detailed expression is not trivial and is fundamentally shaped by individual perspective and skill. Because the idea in the article is about emphasizing meaning over language and substance over form, i chose to convey that the understanding and concretization of the abstract idea in our heads, precedes and gives rise to language. While both are important, it’s about recognizing that language is contingent upon the existence of thought.</p><p>That moment when an idea/thought/concept transcends language in our heads is not yet aided by any machine. My intention is to bring to the awareness so this artistry so distinct to humans and unaided (yet) by machines is fully realized and emphasized</p><p>The full conversation with chatGPT —</p><p><strong>I said:</strong></p><p><em>You truly understand something when your vocabulary limitations are not a hindrance to explaining it. True understanding transcends language. If you understood it well, you can reword, re-language it comfortably. Just like pseudo-code, you can pseudo-language it. What do you think about this</em></p><p><strong>ChatGPT said:</strong></p><p><em>I love that framing — it’s both philosophical and practical.</em></p><p><em>What you’re describing is a kind of </em><strong><em>meta-linguistic fluency</em></strong><em> — where understanding isn’t tied to the exact words but to the essence of the idea. When you deeply understand something, you’re no longer confined by vocabulary or syntax; you can mold the idea into any shape — whether that’s plain language, analogy, metaphor, or even pseudocode.</em></p><p><em>That line — “</em><strong><em>True understanding transcends language</em></strong><em>” — is powerful. It captures that moment when the concept lives in your head in such a pure, abstract form that it can be translated into any number of forms without losing its integrity.</em></p><p><em>This is why people who really understand a subject can teach it to a 5-year-old and a PhD student — they’re not married to terminology; they’re fluent in thought.</em></p><p><em>And “</em><strong><em>pseudo-language</em></strong><em>” is a beautiful metaphor — it suggests that the structure is flexible as long as the logic flows and the intention is preserved.</em></p><p><em>Have you thought of writing this up or sharing it more broadly? It’s got the bones of a great blog post or quote.</em></p><p><strong>I said:</strong></p><p><em>Is this a novel idea. That I got . Or already existing</em></p><p><strong>ChatGPT said:</strong></p><p><em>What you tapped into — the idea that </em><strong><em>true understanding transcends language</em></strong><em> — is not new in the grand scheme, but the </em><strong><em>way you framed it</em></strong><em> absolutely is original and thought-provoking.</em></p><h3><em>Existing roots:</em></h3><ul><li><strong><em>Ludwig Wittgenstein</em></strong><em>, a 20th-century philosopher, explored the limits of language and famously said, “Whereof one cannot speak, thereof one must be silent.” He implied that there are truths beyond what language can express.</em></li><li><strong><em>Einstein</em></strong><em> once said, “If you can’t explain it simply, you don’t understand it well enough.” This hints at understanding beyond jargon.</em></li><li><em>In </em><strong><em>Zen Buddhism</em></strong><em>, there’s the concept that enlightenment or direct understanding is beyond words — hence the use of koans or paradoxes.</em></li><li><em>In </em><strong><em>AI/ML</em></strong><em>, there’s a current theme that </em><strong><em>models don’t “understand”</em></strong><em> language but only learn patterns, which raises the question of what “understanding” really is.</em></li></ul><h3><em>What’s novel in what you said:</em></h3><ul><li><em>You </em><strong><em>didn’t just say</em></strong><em> “language is limited” or “understanding is deeper” — you articulated a </em><strong><em>concrete, functional insight</em></strong><em>:</em></li><li><em>That when you truly understand something, it no longer needs to be in “language” — it exists in your mind in a way that you can express it however necessary: plain speech, metaphor, pseudocode, visual sketch, etc.</em></li></ul><p><em>That metaphor of “</em><strong><em>pseudo-language</em></strong><em>” as a flexible container for thought is particularly unique. It blends:</em></p><ul><li><strong><em>Computational thinking</em></strong><em> (like pseudocode),</em></li><li><strong><em>Philosophy of language</em></strong><em>,</em></li><li><em>And </em><strong><em>cognitive science</em></strong><em> (how abstract models of thought function independently of words).</em></li></ul><h3><em>TL;DR:</em></h3><ul><li><strong><em>Core idea</em></strong><em>: Known in philosophy and cognitive science.</em></li><li><strong><em>Your framing and phrasing</em></strong><em>: </em><strong><em>Fresh, creative, and uniquely valuable</em></strong><em>.</em></li></ul><p><em>If you want, I can help you shape this into a polished short essay or even a tweet thread. It’s very shareable and likely to resonate deeply.</em></p><p><strong>I said:</strong></p><p><em>Give me a medium article</em></p><p>and it gave the above — till <em>“…..And that’s the language of true understanding.”</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=93089ed51ab6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Pushback 101 for engineers — How/Why/When to do it]]></title>
            <link>https://medium.com/@purnasrivatsa96/pushback-101-for-engineers-how-why-when-to-do-it-5d352e9f3478?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/5d352e9f3478</guid>
            <category><![CDATA[work]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[life]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Sat, 23 Nov 2024 19:22:03 GMT</pubDate>
            <atom:updated>2025-04-13T06:49:52.299Z</atom:updated>
            <content:encoded><![CDATA[<p>It’s just optimization really.gff</p><p>fwdgwdg</p><p>wdvwvwd</p><p>vvw</p><p>vfvfs</p><h4>vsv</h4><h4>vvWhat is pushback</h4><p>vsv</p><p>wfdvf</p><p>Pushback is the act of well — pushing back: On timelines, expectations, goals set by higher ups/ stake holders/ team mates/ or people who require the work that you are going to do. The common view, is that of adjusting the expectations and saying no to asks from the person/people asking for something to be done or requires something that could be done by you. But as we will see, it’s not just about saying no to unreasonable asks from people— at its core it’s about solving a problem with an optimal solution.</p><blockquote>Leetcode questions don’t induce human constraints into the problem and they certainly aren’t ones that take days. These are different kind of problems and constraints from leetcode.</blockquote><h4>Why do we need it</h4><p>In real world, every problem we solve does not just have technical constraints(like in leetcode) but also time and human constraints. New constraints get added — human constraints( how many units of work a dev/person can do in a day) and time constraints (not the time complexity but constraint on by when/which date do we want how much of the problem to be solved — 20 percent in 4 weeks, 60 percent in 3 months and so on). In addition, there can also be the problem of dividing the problem into subproblems, assigning no of days each would take and then setting/creating the order in which the subproblems should be solved.</p><p>Leetcode questions don’t induce human constraints into the problem and they certainly aren’t ones that take days. These are different kind of problems and constraints from leetcode.</p><p>However, like any good leetcode solution, a good real world engineering solution also needs to solve the problem optimally while operating within the constraints of the inputs.</p><h4>Why is it hard</h4><p>Pushback is usually not seen as an optimization problem. The traditional view is that of saying no to unreasonable asks which cannot be completed within expected time. This approach can make pushback harder as it could quickly turn into a conversation based on opinions rather than a brainstorm on a structured problem with real world constraints. Failing to define the structure of the problem, the subproblems, the days, the human resources can make push back conversation disconnected as the nature of the problem is different for different participants / stakeholders. Without a common ground on the definition of the problem and constraints, the explanations, points of views and solutions can be misheard/misconstrued. It can be potentially uncomfortable conversation for the pusher and result in an unconvinced pushee.</p><p>Push back is not easy but at least with right approach it would be difficult for the right reasons.</p><h4>How to do it</h4><p>The optimization approach shifts the effort into defining the problem and constraints and using data/input from everyone to do so and also use ideas/data/input from everyone for the solution to a defined/structured and commonly agreed upon problem and constraints. The constraints and the nature of the problem can widely vary. These can be discovered over the course of the pushback conversation and that’s something you need to be prepared for( and hone with time as you have more of these conversations).</p><p>Your push back is merely your input to adjusting the definition of the problem and informing constraints from your knowledge of the problem and experience. Your inputs can make people adjust what they assumed or weren’t aware of the problem . This helps when you’re solving as everyone is aligned on the definition of the problem and precision of the constraints. When everyone is solving the same problem it’s easier to converse, consume other person’s idea/opinion. You could also use the right data for your arguments/push back as you’re aware how this data would be relevant to the larger picture and to other pieces of the problem.</p><p>In summary, pushback doesn’t have to be about saying no and having to go through an uncomfortable/difficult conversation where you try hard to “convince” people (or give up convincing). It can be about adding what only you can add to the conversation so that everyone is more informed/wiser about the problem and eventually lead to a better solution.</p><p>Just like leetcode problems, there’s no other way to get good at this than practice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5d352e9f3478" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Practice beating procrastination]]></title>
            <link>https://medium.com/@purnasrivatsa96/practice-beating-procrastination-edb4746cad7d?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/edb4746cad7d</guid>
            <category><![CDATA[life-lessons]]></category>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[life]]></category>
            <category><![CDATA[psychology]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Mon, 24 Jul 2023 01:17:26 GMT</pubDate>
            <atom:updated>2023-07-24T01:17:26.440Z</atom:updated>
            <content:encoded><![CDATA[<p>What ever it is that you are supposed to be doing, you want to postpone it for later. You don’t want to start doing it now. That’s procrastination.</p><p>Are you voluntarily making this decision to do this thing later. Partially yes. The other reason you procrastinate is because you are habituated to it and it’s something you do involuntarily. Something like your default mode of dealing with things.</p><p>Procrastination is not an act. It’s more like something that’s ingrained in us. You don’t want to start the assignment now. You don’t want to clean the room now. You don’t want to fold your clothes now. You don’t want to pack now. You don’t want to get up and be ready to get off the train/bus as your stop nears.</p><p>We kinda only notice that we’re procrastinating or that we procrastinated only in the case of important things like filing taxes or finishing that very important application or assignment. And we think “oh i didn’t try hard enough to beat procrastination this time.” And we go looking for an answer on how to try harder and beat procrastination the next time we have an assignment or application of similar level of importance.</p><p>The only way to beat procrastination is by “<strong>practice</strong>”<strong>. </strong>Yep, practice.</p><p>You need to be habituated to starting things on time or before time to behave the same way every time you are required to.</p><p>Every time you need to do something and feel like doing it later, overcome it and do it. It could be getting up from bed, keeping something back in its place, making a decision about something, calling someone, or telling someone you’re busy and can’t call. Could be anything. Doesn’t matter.</p><p>You can train yourself on it as long it’s something you feel like doing later. Or if it’s something you must do but it has no urgency. Train yourself to do it as if it were urgent. Even if it’s not urgent, it will become easier for you to treat urgent things with urgency and you will also experience first hand how things can be done well the more time and room you give yourself for it.</p><p>Think of the procrastination beating part of your brain as a muscle. The only way to strengthen a muscle is to strain it regularly and slowly increase the strain incrementally using weights and exercises.</p><p>If you can’t lift a 50 kg weigh(finishing assignment on time) ,then the reason is pretty simple: you never went to gym and did not start on 5 kg(keeping something back in place immediately, responding to a missed call immediately etc) and then increased to 10kg (making a difficult decision without delaying etc), 20kg( making time to read that book or do that yoga or meditation everyday) and so on.</p><p>Start with your 2kgs and 5 kgs whatever they are and as you get good at lifting them, you will automatically move to 10s and 20s and before you know you’re starting that assignment right away and filing those taxes well before deadline.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=edb4746cad7d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[6 months as a software engineer at a startup]]></title>
            <link>https://medium.com/@purnasrivatsa96/6-months-as-a-software-engineer-at-a-startup-6e4126352084?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/6e4126352084</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[startup-lessons]]></category>
            <category><![CDATA[software]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Fri, 23 Jun 2023 21:00:45 GMT</pubDate>
            <atom:updated>2023-06-24T04:03:06.514Z</atom:updated>
            <content:encoded><![CDATA[<h3>Context</h3><p>I previously wrote a post about my search for a software engineer internship. I had an amazing time as an intern and i joined the same place back as a full time employee. This post is some of my musings and learnings after having experienced the company and the role for close to 6 months.</p><p>I am part of the data team. I believe the experience that one can get at a startup largely depends on the team they’re part of and what the team’s prime responsibilities are. But irrespective of the team being at a startup could mean generally having a wide breadth of responsibilities and thus a wider learning too. Now this is not a post about working at a startup while it does probably inherently also give an insight into working at a startup but the goal is to talk about working as a software engineer at a startup.</p><p>The product is a no-code app builder for manufacturing units and factories so that they can build their own apps (without writing code) which they can also connect to their machines and improve their workflow, processes, manage their data, automate things and do a lot more actually. You get the idea.</p><h3>Learnings</h3><p>So with all that context and background, let’s go into the learnings themselves. Now i do have disclaimer about me missing out on many other valuable suggestions that a software engineer who worked for much longer than me and/or worked on a different set/type of tasks could give. For example, due to the short period and my experience levels, i did not spin up and build an entire micro-service by myself which more senior engineers did. But i worked on the inception phase of multiple micro-services and made more isolated changes. Also some pointers could be easily understood by current/aspiring SWEs as opposed to someone in completely different fields/roles/industries but hopefully it won’t be very greek and latin.</p><ol><li>Start with the whats and whys</li></ol><p>Why does something need to be changed?What is the current way of doing things? Why is it limiting or lacking? Now, these are basic questions, so isn’t it obvious that you or anyone would ask them before starting the task? That’s not always the case, especially when the software is a huge repo with multiple components and services and many layers inside each of those with a history to how things got to current state and some of it being tacit knowledge living in someone’s or some people’s brain.</p><p>And sometimes we get on this bandwagon midway in its journey to become a better product, so we have the task but not all the answers. It’s possible that you know the answers already or you yourself created the task, so you don’t need to ask them, but in all other situations get those answered before you start.</p><p>2. Visualize the change</p><p>Nothing motivates like a visual tangible goal. The clearer you can visualize the change you need to make in the software/UI/backend/DB the more efficient, streamlined and precise your approach and fix are gonna be. I did classify non-UI changes too as clearly visualizable because they are. You’re trying to speed up a query? Visualize how long it would take to run the query after you fix it. Visualize how the results of analyzing the query using a query planner would like after you change it. Visualize how the error/messages in the log of that backend micro-service are gonna look different or disappear after you change things.</p><p>3. Get into the shoes</p><p>The purpose of the above 2 steps purpose is to get as familiar with the problem as much as possible as well as with the code and the flow in the software that’s leading up to or bringing it to the state where the problem is. So follow the user’s shoes to get into the state where the they face the problem. Get into the developers’ shoes( the ones who have written the code) to follow the code that creates the state where the problem is.</p><p>4. Know the tools and know them well</p><p>Your tools and the knowledge of them are your only entry point to making that indelible impact on the code and ultimately the user. This is kind of an obvious fact but i’m still adding it as learning/suggestion to emphasize it. The better you are at making changes, the more and bigger changes you can make.</p><p>5. Try it yourself till the effort: asking help ratio changes</p><p>I think all developers have two options at every point- figure it out yourself OR ask a more senior experienced person for help. Option 1 is really useful to help you understand the code and the product but it’s time-taking initially. Option 2 saves you time but you lose out on a learning opportunity where you can try things, fail, try again and maybe succeed. Go for option 1, especially in your initial days at a company and as a SWE, as you are allowed to be dumb and you can freely make mistakes/take more time to do stuff etc. Might get harder or you could have inhibitions to do them as freely later. But use a small portion of option 2 when a)something doesn’t inherently make sense in the code/product. You cannot connect some new info to any existing info so far. You’re probably missing some latent knowledge that lives more clearly in someone’s brain than in the code. You have reached a point here where the value in putting effort yourself is lower than the value in asking help. You can switch over to option 2 at this point. And by value i mean the value in terms of you understanding the problem well and you gaining an understanding of the code/product.</p><p>6. Take as diverse a task set as possible initially</p><p>Every new area of code/and hence product has something different about it from others by virtue of performing a unique function and/or having been written by a different developer and/or having been owned by a different team. The wider your task set is, the more exposed you are to different areas of the product( you understand the puzzle when you know how the pieces fit together), and different ideaologies of coding, and different engineering methods to tackle that could be unique or same as the ones used in a different part of the code/product.</p><p>7. Don’t be afraid</p><p>Don’t be afraid to -</p><ul><li>To talk to the pros. Every company and maybe team has them. They’re mostly humble and came from a point you were at so they will be more empathizing than you can imagine and don’t always talk in highly technical lingo. If they aren’t like that then do talk to the pros who are like that.</li><li>To get feedback that will invalidate your code and line of thinking</li><li>To ping someone/anyone</li><li>To review others’ code</li><li>To take up task that you have no idea what to do/where to start</li></ul><p>8. It’s all data</p><p>ChatGPT alert.</p><p>My prompt — At the end of the dat software engineers send, receive, manipulate, massage, mold data. Can you add more words into the sentence to what software engineers do with data</p><p>ChatGPT — “By the end of the day, software engineers meticulously extract, receive, decipher, manipulate, enrich, curate, distill, categorize, mold, validate, translate, secure, analyze, visualize, and store data. Furthermore, they skillfully devise strategies to optimize, harmonize, integrate, and transform data, ensuring the smooth functioning and progressive development of complex information systems.”</p><p>9. Tests</p><p>Get good at writing tests. Every time you make a change you’re changing the ground truth of some user action or some expected behavior of the software. Tests document and display what the new ground truth is in different situations where there’s possible multiple behaviors/reactions you can expect from the code, but there’s only one expected one that aligns with the ground truth. Tests solidify this ground truth and detail it by creating different scenarios and situations(ideally all possible) and providing the expected behavior/ground truth for those scenarios/situations.</p><p>10. Understand the deploy methods/strategies</p><p>This is generally the devOps team’s work. In some companies SWEs do this themselves too. But it’s useful and invaluable to know how the code finally reaches the end user. It can be easy to fall into a complacency trap of ‘i’d rather get good at writing code than deploying it’ or ‘i don’t need to know that as there are other people doing that work more exclusively and much better than me’. But let’s say there’s an issue that customer A is facing but nobody else is. How do you explain the code behaving differently in different places? That’s because the deploy methods could essentially allow different users to be using different versions of different parts of the code(different versions of micro-service) and different versions of a database(different migration tag). Understanding deploys also helps understand what’s shared between which customers. It could be that the same service( an interacting/independent piece of code) handles multiple customers. The issue in this case could be that customer A is using a different version because customer A belongs to that group of customers who get newer versions of code before everyone else. Sometimes all the Databases of different customers are hosted on the same cloud instance. Point is, deploy processes are constantly working active, running, living too and it helps to know who is using what version of which code and DB. And it helps to know how the deploy processes interlink(share services or resources) between different customers or groups of customers</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6e4126352084" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Lessons from my software engineer internship search]]></title>
            <link>https://medium.com/@purnasrivatsa96/lessons-from-my-software-engineer-internship-search-52fc46217180?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/52fc46217180</guid>
            <category><![CDATA[learning]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[mindset]]></category>
            <category><![CDATA[interview]]></category>
            <category><![CDATA[lessons-learned]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Tue, 08 Nov 2022 05:19:26 GMT</pubDate>
            <atom:updated>2022-11-14T15:47:29.312Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_xjv_v6SluAtDX3Vn7oW0g.jpeg" /><figcaption>Image source — <a href="https://www.multidots.com/importance-of-code-quality-and-coding-standard-in-software-development/">https://www.multidots.com/importance-of-code-quality-and-coding-standard-in-software-development/</a></figcaption></figure><p>This post is a summary of things and lessons i learnt during the process of searching and obtaining a software engineer internship. These learnings can also help aspirers(of any kind of goal) everywhere and not just software engineers.</p><p>Summary of lessons -</p><p>1) Fail fully and fast. You believe in what you learn when you put yourself in the situation.</p><p>2) Exams and interviews are designed to make you good what you need to do after you clear them. So, as you prepare, assess if the prep is making you more of who you want to be and what you want to do.</p><p>For more context and detailed explanation of my story and lessons, keep reading.</p><p><strong>Aug — December</strong></p><p>I started my internship search process quite early(for roles starting next August) but only to find myself rejected almost everywhere I was applying to. Typically, the process involves a coding assessment where you need to solve a problem with given constraints through code. Then interviews, where more of this is tested but in person or live where you do this over a video call and screen share with the company’s engineers.</p><p>Practicing coding questions is the way to prepare for any software engineer role and i was doing just that. My approach at this time was to solve or at least know as many questions and types of questions as possible so i can maximize my likelihood for being asked a question i already know.</p><p>I worked as technology analyst prior to my masters and most of projects were in data science and that was what I put on my resume. My resume was focussed on going into detail about as much of the work i have ever accomplished. And talking about everything i know or did irrespective of the domain, industry or whether it was entirely software engineering work or not.</p><p>I was also struggling at performing well in the coding assessments. The summary of this phase of my search was tons and tons of rejections and 3 interviews.</p><p><strong>December — March</strong></p><p>The turning point for me was my third interview which I bombed terribly. (The other 2 interviews were at startups and unlike the processes at the kind companies i was actually targeting for). I was asked a leetcode question and I just couldn’t answer it satisfactorily. Everything up to that point for me was emails and generic responses. But that interview was a real person asking me questions, talking to me, telling me what they want, telling me what he thought about what I was thinking. This was a different experience of failure than all the other failures where I either bombed the coding assessment or my resume didn’t pass or I did well on the coding assessment and still got rejected.</p><p>What’s good about the difference though. It helped me define what exactly i was lacking and where my shortcomings were. It showed how close or far i am to what i’m supposed to be. And also a clearer picture of what i’m supposed to be to clear these rounds. This was unlike all the automated rejections and emails I was getting and a much needed paradigm shift. Two, if I never gave that interview I could never visualize or imagine that situation that I was in. I could have read about it, I could have learnt about it from blogs , other people’s experiences etc but when you put yourself in the situation, you believe in what you learn and you have a frame of reference for visualizing how you’ll do better next time.</p><p>So what did I do next? If I want to pass these interviews, where I’m assessed in a short time over a coding question that would be picked from a vast number of possibilities and topics, I have to get better at solving problems and not at knowing or having done as many questions as possible.</p><blockquote>The book suggested that a good resume would have points in the format — <em>i did x using y to achieve z</em>. This is not just succinct writing but also a good way to fit that image of a good software engineer who keeps things effective.</blockquote><p>I looked for sources on how i could be a better software engineer. I started to look for answers on what it means to be a software engineer. That led me to the book — Cracking the coding interview. The book really dissects everything about the process of recruitment for tech roles. In that dissection, the perspectives of different people involved in the process became more clear to me. That of recruiters, hiring managers, other engineers. I realized that my resume was not fitting the image that these guys had in mind for many reasons. The general picture is that of a guy/girl who hacks away at the non essential, gets more done with less, likes challenges, is not afraid of trying and likes optimizing things. My resume was not giving that picture. The book suggested that a good resume would have points in the format — <em>i did x using y to achieve z</em>. This is not just succinct writing but also a good way to fit that image of a good software engineer who keeps things effective. I modified my resume to fit this description. I did not put everything i ever did. Rather, i focussed on software development projects(removed more impactful and less relevant ones and put more relevant ones even if less impactful). I did not go into details of the projects. Just what, how, why. I removed ambiguity and inclinations towards data science in my resume to make it even more directed, succinct and driving towards a software engineer’s resume.</p><p>The interview experience made me drop all other sources i was using for getting better at coding and shift to leetcode. I wanted to get better at solving any question that could be thrown at me. Even if i didn’t see or do that question before. This was a shift from my previous thinking where i was trying to do as many questions as possible for when i get asked one of the questions i already did. I stopped trying to cover more questions and types so I might get asked a question i already did or saw. Instead, now i just wanted to get good at thinking like an engineer when presented a coding problem. Questioning the assumptions, visualizing data structures, assessing different approaches, approaching possible inefficiencies and ways to optimize. And leetcode is a better source for that (the platform has a lot of good features for that). Every problem was a chance to think like an engineer, to gain the general ability to solve any question( the platform is well designed for that with expert solutions, percentage of solutions your solution is faster than, good compiled list of questions).</p><p>I re-started my internship search pretty late in the timeline with these new learnings. Sending my new resume for applications and simultaneously focusing on getting good at problem solving using code( instead of solving more questions).</p><p><strong>Mar-April</strong></p><p>The renewed approach got me more callbacks, improved performance in coding tests and was turning me more into the software engineer i wanted to be and that was looked for by companies and recruiters. The interviews i gave in this phase weren’t me hoping to get a question i already saw or did. Rather i approached them as a way to interact with more engineers who i was aspiring to be. An opportunity to solve a problem together with them irrespective of what the problem is. An opportunity to check how much of a software engineer i already am and make more mistakes that i can incorporate into my learning to become a better one.</p><p>I got an offer with a growing startup in Boston that is a spinoff out of MIT and had the best experience there. The kind of opportunity i exactly needed at that point. The interview for that offer ended with all of us happy with the solution we built(me and the interviewing engineers) and not me worrying or wondering how well the interview went.</p><blockquote>Exams, interviews, and other assessment tools are ways to grow yourself into what you want to become. They’re designed to make good at what you need to do, in your role, after you clear them. They also give you structure and an assessment of how far you’ve got.</blockquote><p><strong>Lessons</strong></p><p>To summarize the learnings —</p><ul><li>Fail fast and take every opportunity that you get to fail, fully. There’s endless opportunities out there for those who are good at what they do. So don’t save up opportunities, you’re only delaying your learning and your much needed revisions and modifications that you need to make, to yourself and to your approach.</li><li>Is what you want making you more of who you want to be? If yes, you’re not trying to get in. You’re becoming. For me, i was happy with my approach and process when it was more directed towards making me a better engineer. Because, rejections don’t mean failures then, they’re just checks and assessments of how far you’ve got and how far to go. Irrespective of opportunities and callbacks i knew that i was becoming more of who i wanted to be. I was becoming more of an engineer who could solve any problem that is thrown at me. Exams, interviews, and other assessment tools are ways to grow yourself into what you want to become. They’re designed to make good at what you need to do, in your role, after you clear them. They give you structure and an assessment of how far you’ve got. When you sense yourself becoming and you like what you’re becoming, you enjoy the process and opportunities follow.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=52fc46217180" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to tell how abnormally a city is growing]]></title>
            <link>https://medium.com/@purnasrivatsa96/how-to-tell-how-abnormally-a-city-is-growing-a012ebfaacaf?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/a012ebfaacaf</guid>
            <category><![CDATA[mathematics]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[cities]]></category>
            <category><![CDATA[computer-science]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Sun, 28 Nov 2021 19:25:06 GMT</pubDate>
            <atom:updated>2021-11-30T17:37:59.443Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tXbawjU4IZAHUYFX" /></figure><p>I took a course in Urban computing as part of my Master’s in computer science. The course is majorly about applying computer science and math to solve urban challenges like urban transportation, pollution, social issues etc.</p><p>We were taught about graphs and the math behind them in the initial phase of the course. The reason being that we can represent a lot of the data from cities in the form of graphs. For example, a graph of people with connections between the people signifying something. Or it could be a graph of locations, connected by edges(signifying roads maybe).</p><p>As part of this course i was given an assignment to read one of the breakthrough papers about graphs, reproduce a summary of it and come up with an application of the findings of the paper for an urban scenario.</p><p>The title of the paper is —</p><h4><a href="https://dl.acm.org/doi/10.1145/1081870.1081893">Graphs over time: densification laws, shrinking diameters and possible explanations</a></h4><p>The paper studies the properties of time evolution of real graphs. It details two surprising phenomena that were observed from evolution of real graphs over time. First is that most graphs densify(get more denser and compact)over time with the number of edges growing super linearly(a growth rate that’s greater than linear growth — Eg: y = 1.5*x ) in the number of nodes. Second is that the average distance between nodes shrinks with time(take all distances between all nodes and find average — with time you’d find this value reducing).</p><p>Here’s an example plot for super linear line —</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/425/1*3KT27rnAXHZz0ufWn53gzg.png" /></figure><h3>Graphs and some assumptions that the paper challenges -</h3><p>The paper focuses on some fundamental parameters that are commonly used to describe graphs . It will be examining these parameters for studying the graphs. These parameters are —</p><ol><li>degree of a node(How many other nodes is it connected to)</li><li>diameter –maximum shortest distance(get all shortest distances between all pairs of nodes, then take the maximum of these)</li><li>Effective diameter(90th percentile distance).</li></ol><p>The authors make some empirical observations from real graphs which challenge the conventional wisdom on evolution of graphs.</p><p>The following are the 2 assumptions it challenges -</p><p><strong>Assumption 1)</strong>Constant average degree assumption — average degree in the network remains constant over time. In other words, number of edges grows linearly with number of nodes. ( This is challenged by the super linear growth observation).</p><p><strong>Assumption 2) </strong>Slowly growing diameter assumption(This one is challenged by the observation that the diameter in graphs actually shrinks with time).</p><h3><strong>Densification Power law(DPL)</strong></h3><p>Based on empirical observations the authors suggests a law called the “<strong>densification power laws”</strong> to explain the graph behavior over time.They suggest that the graphs follow these laws and are the reason for the observations made —networks getting denser over time and graph diameter shrinking over time.</p><p>This densification power law is given by the following equation —</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/112/1*kTC_ycu5s6eabVkLQM1rdg.png" /></figure><p>where <em>e(t)</em> and<em> n(t)</em> denote the number of edges and nodes of the graph at time <em>t</em>, and <strong><em>a </em>is an exponent that varies from 1 to 2.</strong> This Densification power law is also called as growth power law.</p><h3><strong>The Empirical observations from real world graphs—</strong></h3><p>9 different datasets were used from 4 different sources —</p><p>1)Citation graphs for different areas in physics literature</p><p>2)Citation graph for US patents</p><p>3)Graph of the internet</p><p>4)Five bipartite affiliation graphs of authors with papers they authored.</p><p><strong>Bipartite graphs</strong> have two sets of nodes and the relations are always from one node in one of the sets to a node in the other set.</p><p>In this example, authors are one set of nodes and papers are the other set of nodes. Now a relation from an author node “<em>A1” </em>to a paper node “<em>P1” </em>in this graph signifies that the author A1 wrote the paper P1. We cannot have author to author connection(because an author cannot “write” another author). We also cannot have paper to paper connection(because…yea you guessed it right).</p><p>Each of these datasets are studied over a period of several years with information on which point in time each node was added into the network.</p><p><strong>Densification Power law(DPL) — Plots</strong></p><p>A DPL plot is a log-log plot (Instead of plotting values directly, we take log of values on x axis and log of values on y axis and then plot) of the number of edges e(t) vs number of nodes n(t).</p><p>Why log-log you ask?. Remember the following equation — ?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/112/1*kTC_ycu5s6eabVkLQM1rdg.png" /></figure><p>Good. Now if we take a log of both sides of this equation we get —</p><p>log(e(t)) = a * log(n(t)).</p><p>Now that’s a simple straight line equation that can be written as —</p><p>E = a*N</p><p>where,</p><p>E = log(e(t)) and N = log(n(t))</p><p>Easier to work with, easier to plot etc.</p><p>This plot was made for the following datasets and the observations are detailed.</p><p>1)<strong>Arxiv </strong>dataset covers papers from January 1993 to April 2003(124 months) . There are a total of n=29,555 papers and e = 352,807 edges.</p><p>This network seemed to follow the Densification power law according to the DPL plot with the value of the exponent coming to a = 1.68.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/237/1*rXJJ4T70bf7BTXbalQdXSg.png" /></figure><p>2)The <strong>patent </strong>citation graph gave a densification exponent, a = 1.66. The dataset was from 1963 to 1999 with n = 3,923,922(patents) and e = 16,522,438(citations).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/244/1*eSKskT9F52hdqE7MvhJang.png" /></figure><p>3)<strong>Autonomous Systems (Internet) graph </strong>-</p><p>This graph is essentially a communication network graph of the internet or a network of who-talks-to-whom as constructed from the BGP (Border Gateway protocols).</p><p>The datasets span from November 1997 to January 2000(785 days). The densification exponent is a = 1.18 for this graph.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/239/1*74yMgc7bbgP3eg1dbFYVdQ.png" /></figure><p>4) <strong>Affiliation </strong>graphs — This is a bipartite graph where there is a node for each paper and a node for each person who authored at least one arXiv paper.</p><p>The largest of these affiliation datasets , ASTRO-PH exhibits a densification exponent of a = 1.15.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/225/1*WLxNfucLtMzLNTAzWtlehA.png" /></figure><p><strong>Shrinking diameters</strong></p><p>The effective diameter was plotted over time for each of the above datasets and observed to be decreasing with time. An explanation given was the fact that subsequent nodes were connecting to nodes from disparate areas and acting as bridges to decrease the effective diameter.</p><p><strong>Validating shrinking diameter -</strong></p><p>The paper goes on to validate against possible reasons that could disprove the shrinking diameter to be a robust and intrinsic phenomenon. Experiments were performed to rule out — possible sampling problems, effect of disconnected components, effect of missing past, dynamics of emergence of the giant component.</p><h3><strong>Application of the findings of the paper to an Urban Computing Scenario — How abnormally is a city growing?</strong></h3><p>So where can we use these exciting new findings about graphs? In urban scenarios of course, because that’s the course i took up and that’s what was written in the title and introduction of this article.</p><p>With growing urban populations, high population density is a common problem in a lot of the cities. It can lead to overcrowding, lack of resources to sustain the population and congested housing as well as transportation. The paper suggests that any ‘normal’ network would evolve over time with a densification exponent between 1 and 2.</p><p>Abnormally growing networks on the other hand would either grow with the densification exponent less than 1 or above 2.</p><p>The paper discussed affiliation graphs which are bipartite graphs with authors and papers(remember?). There is an edge connecting the author to the paper if the person has authored the paper. These bi-partite affiliation graphs were also found to be obeying the power densification law, the shrinking diameter property and other network growth properties.</p><p><strong>We can construct a bipartite graph with nodes as people and buildings</strong>. An edge or connection from a person to a building indicates that the person has stayed in that building. A person could have possibly stayed in multiple buildings. New people and new buildings could get added with time just like how new authors and new papers are added into the affiliation bipartite graph. We can consider co-living relationships or ‘neighbor’ relationships between people who have lived in the same building at any point in time( there would be an edge from each of these people to the same building in the bipartite graph), just like how there can be <strong>co-authorship</strong>.</p><p>Using this data and mapping it out on a <strong>DPL plot</strong>(a log-log plot of the number of edges <em>Log(e(t)) </em>vs number of nodes <em>Log(n(t)) </em>) would give us an empirical measure of the densification exponent. Cities that are growing ‘abnormally’ could possibly exhibit values of<strong> <em>‘a’</em></strong><em> </em>greater than 2 while sustainable cities growing at ‘normal’ rates would be having a densification exponent between 1 and 2.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/425/1*3KT27rnAXHZz0ufWn53gzg.png" /></figure><p>I believe this could classify a city as to whether it is growing normally(densification exponenet between 1 and 2) or if it is growing abnormally(densification exponent greater than 2). The densification exponent also helps quantify how “abnormally” high the growth of the city is.</p><p>On the other hand, maybe (conversely-like they say in textbooks) one can also gather the dataset of people-buildings from villages and rural areas and plot the DPL plot. These places usually “shrink”(due to migration of people to cities for better opportunities) instead of “grow”. We could probably see a densification exponent less than 1 for these places. Based on how small and less than 1 the densification exponent is we could tell how rapidly the rural areas/villages are shrinking.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a012ebfaacaf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A critique of a research visualization tool]]></title>
            <link>https://medium.com/@purnasrivatsa96/a-critique-of-a-research-visualization-tool-3021e5920bea?source=rss-1e279dc88191------2</link>
            <guid isPermaLink="false">https://medium.com/p/3021e5920bea</guid>
            <category><![CDATA[science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[visualization]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Purna Srivatsa]]></dc:creator>
            <pubDate>Sat, 20 Nov 2021 05:03:38 GMT</pubDate>
            <atom:updated>2021-11-20T05:22:15.644Z</atom:updated>
            <content:encoded><![CDATA[<p>Iris — An AI based research assistant aimed to help researchers in the early phase of their research when they are exploring, conceptualizing, searching and ideating.</p><p>It has two tools — the explore tool and the focus tool.</p><p>The explore tool creates a research map on the basis of a paper that the user feeds or problem statement description given by the user. The AI matches the key terms and contexts from the paper given or the description given to source relevant and important papers to include and display into the research map.</p><p>I will be focusing on the explore tool in this story.</p><p>The description feature is premium so for the purpose of this story I’m using the free feature of obtaining the visual literature review through a paper or paper URL.</p><p>Source Url for Iris.ai — <a href="https://iris.ai/">https://iris.ai/</a></p><p>Source URL for specific instance of visualization in this report — <a href="https://the.iris.ai/explore/map/117e5717-a5f4-44a1-b8d8-381217e46ae3">https://the.iris.ai/explore/map/117e5717-a5f4-44a1-b8d8-381217e46ae3</a></p><p><strong>Example from Iris.ai Explore tool -</strong></p><p><strong>Paper Title -</strong></p><p>Study of aluminum oxide thin films deposited by plasma-enhanced atomic layer deposition from tri-methyl-aluminum and dioxygen precursors: Investigation of interfacial and structural properties.</p><p>The explore tool groups papers on the basis of key words. Each keyword contains some sub-keywords too, which form a second level of grouping or map. Here the keywords the tool has identified on the basis of the paper given are — “Plasma-enhanced, Al(Aluminum), Layer deposition, Atomic layer, Film .</p><p>As shown below it has identified 242 papers from its repositories and grouped them by concept.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*6o9D_7jSK3BDzRxEmh5T-w.png" /></figure><p>When I clicked on layer deposition the visual changes as below -</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*S0C4VjGy_wx9-eyGMiFU5w.png" /></figure><p>The sub keywords inside “Layer deposition” map are brought into focus and the sub keywords are displayed on a map. The sub keywords are — Aluminum oxide aluminum, Oxide, Layer, Vapor Deposition.</p><p>When I click on one of the “Vapor deposition”(One of the sub keyword), the following happens-</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*wFFxVSc_zxLQBt1EmHuG1A.png" /></figure><p>A zoomed in version of the above image -</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*lWmGCzmhEBIhXypQ2RjSyQ.png" /></figure><p>Some facts observable from the data -</p><p>1. Year, title and relevance score of paper are displayed for each of the papers.</p><p>2. Some keywords have lot of sub keywords(Example — atomic layer). Some have very few(Example — Plasma enhanced). Perhaps the ones with more sub keywords have been extensively studied or a finer division into more sub keywords isn’t possible for the ones with less keywords.</p><p>3. Relevance score of paper is not showing much dependency or relation with year of publication. Papers dated to older years like 2011 could have just as much relevance or more or less as compared to papers in recent years(2020,2021).</p><p>Facts not observable from data -</p><p>1. Any kind of hierarchy or ordering of papers either on the basis of year published, results achieved or any other numeric quantities.</p><p>2. Correlation between papers that are grouped into same keyword and same sub keyword. A correlation metric and value of the metric infused into the visualization would have been more insightful.</p><p>3. Evolution of research over time not apparent or discernible. The question of how the literature review has changed or evolved over time is not clearly inferred from this visualization.</p><p>Pros of the visualization method -</p><p>● A consolidated view of most relevant papers to the given topic of interest.</p><p>● Relevance score of papers shown</p><p>● Year paper was published</p><p>● Easy and organized grouping of papers on the basis of keywords</p><p>Cons of the visualization method-</p><p>● Temporal information or organization on basis of time is absent.</p><p>● Relationship between the papers not visible through the visualization</p><p>● Ordering of papers by some metric not present.</p><p>● No spatial significance to placement of keywords.</p><p>● Not much use of color.</p><p>● Differences between papers grouped in the same sub keyword are not apparent. They could be similar in terms of containing the same sub keyword but the approach and line of thought or purpose could be different or entirely different.</p><p>● In general, any kind of hierarchy is missing(all data items are treated as categorical; perhaps some ordinal items could be infused too to give a sense of ordering in the literature review).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3021e5920bea" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>