<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Srijan Saket — Writing</title>
    <link>https://srijansaket.com/writing/</link>
    <atom:link href="https://srijansaket.com/rss.xml" rel="self" type="application/rss+xml" />
    <description>Essays on retrieval, agent systems, and evaluation.</description>
    <language>en-us</language>
    <lastBuildDate>Sat, 19 Sep 2026 05:40:53 GMT</lastBuildDate>
    <item>
      <title>Context Aging: What an Agent Forgets While It Is Still Running</title>
      <link>https://srijansaket.com/writing/context-aging/</link>
      <guid isPermaLink="true">https://srijansaket.com/writing/context-aging/</guid>
      <pubDate>Mon, 14 Sep 2026 00:00:00 GMT</pubDate>
      <description>An agent read a pricing rule with two constant eras at tool call 12 — correctly, aloud, in its own reasoning. Two hundred and eighty calls later, the document it finished described the rule with one constant. Nothing had been truncated; the fact was in the context the whole time. Context rot, lost-in-the-middle and context poisoning are about how much, where, and what. This is about when — and why the fix is a durable tier the agent writes to at the moment of observation, reads back before it drafts, and is checked against by code before it is allowed to say it is done.</description>
      <author>Srijan Saket with Susnato Dhar</author>
      <category>agents</category>
      <category>context-engineering</category>
      <category>memory</category>
    </item>
    <item>
      <title>Agent Context Is a Recommendation Problem</title>
      <link>https://srijansaket.com/writing/agent-context-is-a-recommendation-problem/</link>
      <guid isPermaLink="true">https://srijansaket.com/writing/agent-context-is-a-recommendation-problem/</guid>
      <pubDate>Tue, 08 Sep 2026 00:00:00 GMT</pubDate>
      <description>We built the retrieval system, registered the tools, wrote careful descriptions, and the agent never called them. Not once. The instinct was to write better descriptions. The correct diagnosis was that we had built a catalogue and expected it to behave like a recommendation — and the difference between those two things is most of what recommender systems research has been about since the mid-2000s.</description>
      <author>Srijan Saket</author>
      <category>retrieval</category>
      <category>agents</category>
      <category>recsys</category>
      <category>context-engineering</category>
    </item>
    <item>
      <title>Your Eval Has a Noise Floor</title>
      <link>https://srijansaket.com/writing/your-eval-has-a-noise-floor/</link>
      <guid isPermaLink="true">https://srijansaket.com/writing/your-eval-has-a-noise-floor/</guid>
      <pubDate>Sat, 05 Sep 2026 00:00:00 GMT</pubDate>
      <description>Re-judging a byte-identical artifact — same text, same rubric, same judge, one more pass — moved its score by four claims in sixty. Re-running an unchanged configuration moved it by about six. Which means any single-run improvement smaller than six claims was indistinguishable from running the same thing twice. Here is how to find your own floor, and what it costs.</description>
      <author>Srijan Saket</author>
      <category>evaluation</category>
      <category>benchmarks</category>
      <category>agents</category>
    </item>
    <item>
      <title>Completion Is a Proof, Not a Status</title>
      <link>https://srijansaket.com/writing/completion-is-a-proof/</link>
      <guid isPermaLink="true">https://srijansaket.com/writing/completion-is-a-proof/</guid>
      <pubDate>Sat, 22 Aug 2026 00:00:00 GMT</pubDate>
      <description>A command in a remote sandbox wrote one unbroken 4 MiB line. The HTTP stream returned no output and no terminal event, the producer was still alive blocked on its pipe, and the request had already discarded the handle used to cancel it. To the caller the tool call was over. To the machine the computation had not ended. Four independent clocks settle at the end of an agent run, and most platforms compress all four into a single status field.</description>
      <author>Srijan Saket with Vivek Sharma</author>
      <category>agents</category>
      <category>systems</category>
      <category>reliability</category>
    </item>
    <item>
      <title>Retrieval Is Not Delivery</title>
      <link>https://srijansaket.com/writing/retrieval-is-not-delivery/</link>
      <guid isPermaLink="true">https://srijansaket.com/writing/retrieval-is-not-delivery/</guid>
      <pubDate>Sun, 02 Aug 2026 00:00:00 GMT</pubDate>
      <description>At baseline, a two-agent question-answering workflow found 65% of the expected facts while reading — but only 22% of that evidence survived the handoff to the writer, and 23% reached the final answer. Retrieval was never the bottleneck. The handoff was. Eight versions and 45 fixed questions later, here is what it took to measure the whole path instead of the last step.</description>
      <author>Srijan Saket with Susnato Dhar</author>
      <category>retrieval</category>
      <category>agents</category>
      <category>evaluation</category>
    </item>
    <item>
      <title>Crash-Proof Custom Agents</title>
      <link>https://bitsnneurons.substack.com/p/crash-proof-custom-agents</link>
      <guid isPermaLink="true">https://bitsnneurons.substack.com/p/crash-proof-custom-agents</guid>
      <pubDate>Sat, 07 Mar 2026 00:00:00 GMT</pubDate>
      <description>Orchestrating long-running AI agents is a distributed-systems problem, not a model-intelligence problem. Our first pass — Celery plus database flags — split-brained on every crash and cancel. The fix was a durable-execution plane (Temporal), coarse activity boundaries, and signal-based human-in-the-loop resumption.</description>
      <author>Srijan Saket</author>

    </item>
    <item>
      <title>BYOC is not a Helm chart: Threat models and hard lessons</title>
      <link>https://bitsnneurons.substack.com/p/byoc-is-not-a-helm-chart-threat-models</link>
      <guid isPermaLink="true">https://bitsnneurons.substack.com/p/byoc-is-not-a-helm-chart-threat-models</guid>
      <pubDate>Wed, 28 Jan 2026 00:00:00 GMT</pubDate>
      <description>“Can you run this in our environment?” hides a stack of security and ops requirements that don’t exist in SaaS: the buyer controls the runtime, the network, the databases, and can often inspect every artifact you ship. A walk through the threat model, the gaps a Helm chart leaves behind, and what it actually takes to ship bring-your-own-cloud safely.</description>
      <author>Srijan Saket</author>

    </item>
    <item>
      <title>Optimizing Golang Jobs for Cost Efficiency: A Beginner&apos;s Guide</title>
      <link>https://bitsnneurons.substack.com/p/optimizing-golang-jobs-for-cost-efficiency</link>
      <guid isPermaLink="true">https://bitsnneurons.substack.com/p/optimizing-golang-jobs-for-cost-efficiency</guid>
      <pubDate>Tue, 27 Aug 2024 00:00:00 GMT</pubDate>
      <description>A practical field guide to cutting Golang workload cost: GOMEMLIMIT, GOGC, GOMAXPROCS tuning, CPU profiling, and Kubernetes resource-request hygiene. Small, well-placed changes that add up to material line-item savings on GKE.</description>
      <author>Srijan Saket</author>

    </item>
    <item>
      <title>Recommended — Just For You: How We Handle Personalisation at Scale</title>
      <link>https://sharechat.com/blogs/artificial-intelligence/recommended-just-for-you-how-we-handle-personalisation-at-scale</link>
      <guid isPermaLink="true">https://sharechat.com/blogs/artificial-intelligence/recommended-just-for-you-how-we-handle-personalisation-at-scale</guid>
      <pubDate>Tue, 01 Nov 2022 00:00:00 GMT</pubDate>
      <description>How ML powers the home feed on ShareChat and Moj at 200M-user scale.</description>
      <author>Srijan Saket</author>

    </item>
    <item>
      <title>Multimodal Automated Content Moderation: Preserving Integrity at ShareChat &amp; Moj (Part I)</title>
      <link>https://medium.com/sharechat-techbyte/multimodal-automated-content-moderation-69876e6a9d85</link>
      <guid isPermaLink="true">https://medium.com/sharechat-techbyte/multimodal-automated-content-moderation-69876e6a9d85</guid>
      <pubDate>Mon, 01 Aug 2022 00:00:00 GMT</pubDate>
      <description>Multi-modal approaches to automated content moderation — combining text, image, and audio signals.</description>
      <author>Srijan Saket</author>

    </item>
    <item>
      <title>30 Hours — The Launch of Moj</title>
      <link>https://sharechat.com/blogs/product/30-hours-the-launch-of-moj</link>
      <guid isPermaLink="true">https://sharechat.com/blogs/product/30-hours-the-launch-of-moj</guid>
      <pubDate>Fri, 01 Jul 2022 00:00:00 GMT</pubDate>
      <description>Behind the scenes of building and launching Moj in 30 hours on July 1, 2020 — and the two-year journey that followed.</description>
      <author>Srijan Saket</author>

    </item>
    <item>
      <title>Fast, Accurate and Scalable Video Content Moderation</title>
      <link>https://sharechat.com/blogs/data-science/fast-accurate-and-scalable-video-content-moderation</link>
      <guid isPermaLink="true">https://sharechat.com/blogs/data-science/fast-accurate-and-scalable-video-content-moderation</guid>
      <pubDate>Sun, 01 May 2022 00:00:00 GMT</pubDate>
      <description>Spatial + temporal ML for moderating short-form video at scale.</description>
      <author>Srijan Saket</author>

    </item>
    <item>
      <title>#LifeAtShareChat — Srijan Saket — Humming In Hall 3 to Crooning In A Studio</title>
      <link>https://sharechat.com/blogs/life-at-sharechat/sharechatinsider-srijan-saket-humming-in-hall-3-to-crooning-in-a-studio</link>
      <guid isPermaLink="true">https://sharechat.com/blogs/life-at-sharechat/sharechatinsider-srijan-saket-humming-in-hall-3-to-crooning-in-a-studio</guid>
      <pubDate>Tue, 01 Jun 2021 00:00:00 GMT</pubDate>
      <description>A personal profile on growing up in a musical family and finding the parallels between music and engineering.</description>
      <author>Srijan Saket</author>

    </item>
  </channel>
</rss>
