<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Ilya Manyakin — Notes</title>
    <link>https://stolas.lol/en/notes/</link>
    <guid isPermaLink="true">https://stolas.lol/en/notes/</guid>
    <description>Writing on Python, Kubernetes and ML in production.</description>
    <language>en</language>
    <lastBuildDate>Wed, 12 Aug 2026 00:00:00 GMT</lastBuildDate>
    <atom:link href="https://stolas.lol/en/notes/rss.xml" rel="self" type="application/rss+xml"/>
    <item>
      <title>Rollback matters more than a perfect deploy</title>
      <link>https://stolas.lol/en/notes/rollback-first/</link>
      <guid isPermaLink="true">https://stolas.lol/en/notes/rollback-first/</guid>
      <description>The perfect deploy happens once a year; every system falls over eventually. That is why the first scenario our CI tests is not the rollout but the rollback.</description>
      <pubDate>Wed, 12 Aug 2026 00:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Models live in production, not on a laptop</title>
      <link>https://stolas.lol/en/notes/model-lives-in-prod/</link>
      <guid isPermaLink="true">https://stolas.lol/en/notes/model-lives-in-prod/</guid>
      <description>A laptop with 97% accuracy and a service running the same checkpoint are two different models. What changes when inference moves into Kubernetes.</description>
      <pubDate>Thu, 04 Jun 2026 00:00:00 GMT</pubDate>
    </item>
    <item>
      <title>The boring layer in front of the model</title>
      <link>https://stolas.lol/en/notes/boring-gateway/</link>
      <guid isPermaLink="true">https://stolas.lol/en/notes/boring-gateway/</guid>
      <description>Half of all &quot;inference problems&quot; are solved by an nginx in front of FastAPI: limits, timeouts, caching and honest 503s. Cheaper than scaling GPUs.</description>
      <pubDate>Thu, 19 Mar 2026 00:00:00 GMT</pubDate>
    </item>
  </channel>
</rss>
