<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Writing by Abhisar Mehta</title>
    <link>https://abhisarmehta.com/writing</link>
    <description>Technical reports about adaptive neural networks, ML observability, and browser inference.</description>
    <language>en</language>
    <atom:link href="https://abhisarmehta.com/feed.xml" rel="self" type="application/rss+xml"/>
    <item><title>I could log every architectural event. I still couldn’t see the model.</title><link>https://abhisarmehta.com/stories/seeing-model-change</link><guid isPermaLink="true">https://abhisarmehta.com/stories/seeing-model-change</guid><pubDate>Sun, 12 Apr 2026 00:00:00 GMT</pubDate><description>How architecture logs became an interactive topology, capacity, importance, and stability viewer.</description></item>
    <item><title>What happens when a transformer is allowed to redesign itself?</title><link>https://abhisarmehta.com/stories/neuroplastic-transformer</link><guid isPermaLink="true">https://abhisarmehta.com/stories/neuroplastic-transformer</guid><pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate><description>A source-backed report on 236 structural decisions across a 30,000-step training run.</description></item>
    <item><title>Training a voice for $14, then fitting it inside the browser</title><link>https://abhisarmehta.com/stories/browser-voice</link><guid isPermaLink="true">https://abhisarmehta.com/stories/browser-voice</guid><pubDate>Wed, 18 Feb 2026 00:00:00 GMT</pubDate><description>VITS training, ONNX export, quantization, and WebAssembly deployment without an inference server.</description></item>
  </channel>
</rss>
