<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en">
  <title>Sharad Bapat</title>
  <subtitle>Experiments and writing.</subtitle>
  <id>https://sharadbapat.com/</id>
  <link rel="alternate" type="text/html" href="https://sharadbapat.com/"/>
  <link rel="self" type="application/atom+xml" href="https://sharadbapat.com/feed.xml"/>
  <updated>2026-09-25T00:00:00Z</updated>
  <author><name>Sharad Bapat</name><uri>https://sharadbapat.com/</uri></author>
  <icon>https://sharadbapat.com/favicon.svg</icon>

  <entry>
    <title>Every cover here is CSS</title>
    <id>https://sharadbapat.com/writing/every-cover-here-is-css/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/writing/every-cover-here-is-css/"/>
    <published>2026-09-25T00:00:00Z</published>
    <updated>2026-09-25T00:00:00Z</updated>
    <category term="writing"/>
    <summary>The moving covers on this site are painted by the browser from eight CSS recipes and ten palettes with Hindi names. No images, no JavaScript.</summary>
  </entry>

  <entry>
    <title>Interactive lessons</title>
    <id>https://sharadbapat.com/experiments/lessons/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/lessons/"/>
    <published>2026-09-25T00:00:00Z</published>
    <updated>2026-09-25T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>Three five-minute lessons where you predict first and find out second: Monty Hall, Simpson's paradox and why √2 is not a fraction.</summary>
  </entry>

  <entry>
    <title>Scan or text?</title>
    <id>https://sharadbapat.com/experiments/scan-or-text/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/scan-or-text/"/>
    <published>2026-09-25T00:00:00Z</published>
    <updated>2026-09-25T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>A 110 KB WebAssembly classifier that decides whether a PDF needs OCR in about a millisecond, without rendering it. 525 of 525 right. Try it here.</summary>
  </entry>

  <entry>
    <title>Brief</title>
    <id>https://sharadbapat.com/experiments/brief/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/brief/"/>
    <published>2026-09-24T00:00:00Z</published>
    <updated>2026-09-24T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>A calm, newest-first reader that merges RSS, Atom and JSON feeds into one queue and opens the full article, running on Cloudflare Workers.</summary>
  </entry>

  <entry>
    <title>Before History, Chapter 0</title>
    <id>https://sharadbapat.com/experiments/before-history/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/before-history/"/>
    <published>2026-09-23T00:00:00Z</published>
    <updated>2026-09-24T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>A 67-second narrated documentary about India's deep past, drawn with Canvas and scored with Web Audio. Every frame and every sound is code.</summary>
  </entry>

  <entry>
    <title>Jev eval</title>
    <id>https://sharadbapat.com/experiments/jev-eval/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/jev-eval/"/>
    <published>2026-09-22T00:00:00Z</published>
    <updated>2026-09-24T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>An evaluation of TypeSafe's Jev judgment model: 97.7% accurate, with one real flaw and a one-line fix. Give the model the rule, not just the request.</summary>
  </entry>

  <entry>
    <title>Latch</title>
    <id>https://sharadbapat.com/experiments/latch/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/latch/"/>
    <published>2026-09-20T00:00:00Z</published>
    <updated>2026-09-24T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>Can a model reword text while provably keeping its facts? Seven verifier designs and two blind tests said not reliably, so I stopped.</summary>
  </entry>

  <entry>
    <title>Blind spreadsheet benchmark</title>
    <id>https://sharadbapat.com/experiments/blind-spreadsheet-benchmark/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/blind-spreadsheet-benchmark/"/>
    <published>2026-09-13T00:00:00Z</published>
    <updated>2026-09-25T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>Twenty real UK public spreadsheets, three editions each, labelled for what changed between editions. In a third of cases, a careful reader should stop.</summary>
  </entry>

  <entry>
    <title>NDA clause model</title>
    <id>https://sharadbapat.com/experiments/nda-clause-model/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/nda-clause-model/"/>
    <published>2026-09-04T00:00:00Z</published>
    <updated>2026-09-25T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>A 184M-parameter model that reads an NDA, rules on 17 standard clauses and quotes its evidence. Tested once on a sealed set: 84.5% accuracy.</summary>
  </entry>

  <entry>
    <title>ALTXI</title>
    <id>https://sharadbapat.com/experiments/altxi/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/experiments/altxi/"/>
    <published>2026-08-23T00:00:00Z</published>
    <updated>2026-09-25T00:00:00Z</updated>
    <category term="experiment"/>
    <summary>Cricket between people you know, played out by a seeded simulator. The result is fixed before the first ball, and your browser can check it.</summary>
  </entry>

  <entry>
    <title>The best AI strategy is sometimes no AI</title>
    <id>https://sharadbapat.com/writing/the-best-ai-strategy-is-no-ai/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/writing/the-best-ai-strategy-is-no-ai/"/>
    <published>2026-08-19T00:00:00Z</published>
    <updated>2026-09-24T00:00:00Z</updated>
    <category term="writing"/>
    <summary>A client wanted AI in their workflow. I told them not to: the problem was deterministic, and intelligence would only have added cost, latency and doubt.</summary>
  </entry>

  <entry>
    <title>A typing animation with zero JavaScript</title>
    <id>https://sharadbapat.com/writing/pure-css-typewriter/</id>
    <link rel="alternate" type="text/html" href="https://sharadbapat.com/writing/pure-css-typewriter/"/>
    <published>2026-08-16T00:00:00Z</published>
    <updated>2026-09-24T00:00:00Z</updated>
    <category term="writing"/>
    <summary>A typewriter effect that types, pauses, deletes and starts again, built entirely in CSS. No JavaScript, no timers, no libraries.</summary>
  </entry>
</feed>
