<?xml version="1.0" encoding="UTF-8" ?>
<rss version="2.0">
    <channel>
      <title>Rituraj</title>
      <link>https://rituraj003.github.io</link>
      <description>Last 10 notes on Rituraj</description>
      <generator>Rituraj</generator>
      <item>
    <title>~/blog</title>
    <link>https://rituraj003.github.io/blog/</link>
    <guid>https://rituraj003.github.io/blog/</guid>
    <description><![CDATA[ Working notes, essays, and field reports from strategy + systems work. ]]></description>
    <pubDate>Thu, 23 Jul 2026 13:50:40 GMT</pubDate>
  </item><item>
    <title>Rituraj</title>
    <link>https://rituraj003.github.io/</link>
    <guid>https://rituraj003.github.io/</guid>
    <description><![CDATA[ Verifier-guided inference, evaluation systems, and research on long-horizon reasoning. ]]></description>
    <pubDate>Thu, 23 Jul 2026 13:50:40 GMT</pubDate>
  </item><item>
    <title>~/now</title>
    <link>https://rituraj003.github.io/now/</link>
    <guid>https://rituraj003.github.io/now/</guid>
    <description><![CDATA[  Building a policy-repair stack that combines [prm, search, eval] for high-stakes reasoning workflows. ]]></description>
    <pubDate>Thu, 23 Jul 2026 13:50:40 GMT</pubDate>
  </item><item>
    <title>~/projects</title>
    <link>https://rituraj003.github.io/projects/</link>
    <guid>https://rituraj003.github.io/projects/</guid>
    <description><![CDATA[ Systems I am building at the intersection of reasoning quality, evaluation, and infrastructure. ]]></description>
    <pubDate>Thu, 23 Jul 2026 13:50:40 GMT</pubDate>
  </item><item>
    <title>~/research</title>
    <link>https://rituraj003.github.io/research/</link>
    <guid>https://rituraj003.github.io/research/</guid>
    <description><![CDATA[ Papers and preprints I revisit while building systems. Each entry has a five-line strategy summary. ]]></description>
    <pubDate>Thu, 23 Jul 2026 13:50:40 GMT</pubDate>
  </item><item>
    <title>The Response Curve</title>
    <link>https://rituraj003.github.io/the-response-curve</link>
    <guid>https://rituraj003.github.io/the-response-curve</guid>
    <description><![CDATA[ What the behavior-vs-strength curve looks like when you steer a model, and which one-pass screens actually predict it. ]]></description>
    <pubDate>Fri, 17 Jul 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Dense Supervision Is Not Enough: The Readout Blind Spot in Looped Language Models</title>
    <link>https://rituraj003.github.io/research/readout-blind-spot</link>
    <guid>https://rituraj003.github.io/research/readout-blind-spot</guid>
    <description><![CDATA[ Rituraj Sharma · Tu Vu — Virginia Tech · preprint, 2026 PDF Code arXiv · pending Abstract Looped language models turn hidden states into runtime state: each state is decoded for prediction and fed back into future computation. ]]></description>
    <pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>ghostenv: Stop Leaking Secrets to AI Agents</title>
    <link>https://rituraj003.github.io/blog/ghostenv-stop-leaking-secrets-to-ai-agents</link>
    <guid>https://rituraj003.github.io/blog/ghostenv-stop-leaking-secrets-to-ai-agents</guid>
    <description><![CDATA[ Your .env file is the first thing an AI agent reads. It has to. That’s where the config lives. ]]></description>
    <pubDate>Wed, 22 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment</title>
    <link>https://rituraj003.github.io/research/master-key-hypothesis</link>
    <guid>https://rituraj003.github.io/research/master-key-hypothesis</guid>
    <description><![CDATA[ Rishab Balasubramanian · Pin-Jie Lin · Rituraj Sharma · Anjie Fang · Fardin Abdi · Viktor Rozgic · Zheng Du · Mohit Bansal · Tu Vu arXiv PDF Abstract We investigate whether post-trained capabilities can be transferred across models without retraining, with a focus on transfer across different model ... ]]></description>
    <pubDate>Tue, 07 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>PRISM: Pushing the Frontier of Deep Think via Process Reward Model-Guided Inference</title>
    <link>https://rituraj003.github.io/research/prism-prm-guided-inference-for-deep-think</link>
    <guid>https://rituraj003.github.io/research/prism-prm-guided-inference-for-deep-think</guid>
    <description><![CDATA[ Rituraj Sharma, Weiyuan Chen, Noah Provenzano, Tu Vu · arXiv preprint · 2026 arXiv PDF Code Abstract PRISM uses step-level verification (a PRM signal) to guide population refinement and solution aggregation in Deep Think systems, treating candidates like particles in an energy landscape. ]]></description>
    <pubDate>Tue, 03 Mar 2026 00:00:00 GMT</pubDate>
  </item>
    </channel>
  </rss>