<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Himanshu - Principal AI Specialist</title><description>Himanshu - Principal AI Specialist in Bengaluru. I build GenAI and agentic systems for pharma, run small experiments on whether they actually work, and write about both. Occasionally I walk up mountains.</description><link>https://bayesianhimanshu.org/</link><language>en</language><item><title>System Design Isn&apos;t Changing. The System Is.</title><link>https://bayesianhimanshu.org/blog/personal/system-design-changing/</link><guid isPermaLink="true">https://bayesianhimanshu.org/blog/personal/system-design-changing/</guid><description>On agent harnesses, context engineering, code agents, and a question that gets asked in a way that guarantees a bad answer</description><pubDate>Thu, 20 Aug 2026 00:00:00 GMT</pubDate><category>personal</category><category>system-design</category><category>agents</category><category>architecture</category><category>methodology</category></item><item><title>I Built the Same Agent Twice to Find Out If the Harness Matters</title><link>https://bayesianhimanshu.org/blog/professional/harness-lab/</link><guid isPermaLink="true">https://bayesianhimanshu.org/blog/professional/harness-lab/</guid><description>A controlled comparison of a hand-rolled Python harness and Google ADK - and what happened when I audited my own headline</description><pubDate>Thu, 20 Aug 2026 00:00:00 GMT</pubDate><category>professional</category><category>agents</category><category>evaluation</category><category>adk</category><category>pharmacovigilance</category><category>methodology</category></item><item><title>The Life and Lies of Data Contextualization for Agentic AI</title><link>https://bayesianhimanshu.org/blog/professional/the-life-and-lies-of-data-contextualization/</link><guid isPermaLink="true">https://bayesianhimanshu.org/blog/professional/the-life-and-lies-of-data-contextualization/</guid><description>Does your AI agent actually know what it doesn&apos;t know? 3,800 controlled experiments across three frontier models, measuring whether the metadata around your data changes how reliable the agent is.</description><pubDate>Mon, 10 Aug 2026 00:00:00 GMT</pubDate><category>professional</category><category>evaluation</category><category>agents</category><category>metadata</category><category>reliability</category></item><item><title>When the Data Refuses to Behave: Building an Agentic Demand Forecasting System</title><link>https://bayesianhimanshu.org/blog/professional/when-the-data-refuses-to-behave/</link><guid isPermaLink="true">https://bayesianhimanshu.org/blog/professional/when-the-data-refuses-to-behave/</guid><description>How classical time series, deep learning, and Generative AI come together to tackle one of supply chain&apos;s oldest problems — built on the M5 competition data, where 68% of item-level observations are zero.</description><pubDate>Mon, 10 Aug 2026 00:00:00 GMT</pubDate><category>professional</category><category>forecasting</category><category>time-series</category><category>agents</category><category>supply-chain</category><category>m5</category></item><item><title>Pilgrimage to Kedarnath: A Child&apos;s Manifestation or Faith?</title><link>https://bayesianhimanshu.org/blog/personal/kedarnath/</link><guid isPermaLink="true">https://bayesianhimanshu.org/blog/personal/kedarnath/</guid><description>Twenty-two kilometres up to a temple I had been thinking about since I was a child, and the question I could not settle on the way.</description><pubDate>Sun, 09 Aug 2026 00:00:00 GMT</pubDate><category>personal</category><category>kedarnath</category><category>travel</category><category>trek</category></item><item><title>केदारनाथ की तीर्थयात्रा: एक बच्चे की कल्पना या आस्था?</title><link>https://bayesianhimanshu.org/blog/personal/kedarnath-hind/</link><guid isPermaLink="true">https://bayesianhimanshu.org/blog/personal/kedarnath-hind/</guid><description>एक ऐसे मंदिर तक बाईस किलोमीटर की चढ़ाई जिसके बारे में मैं बचपन से सोचता आ रहा था, और वह सवाल जिसे मैं पूरे रास्ते सुलझा नहीं पाया।</description><pubDate>Sun, 09 Aug 2026 00:00:00 GMT</pubDate><category>personal</category><category>kedarnath</category><category>travel</category><category>trek</category></item><item><title>Project VIGIL - Agentic Pharmacovigilance</title><link>https://bayesianhimanshu.org/blog/professional/project-vigil/</link><guid isPermaLink="true">https://bayesianhimanshu.org/blog/professional/project-vigil/</guid><description>Agentic Pharmacovigilance - automating adverse-event case processing and safety signal detection</description><pubDate>Fri, 03 Jul 2026 00:00:00 GMT</pubDate><category>professional</category><category>Pharmacovigilance</category><category>agents</category><category>gcp</category></item><item><title>VERDICT: Validation and Explainability framework for Reproducible, Deterministic, Inspectable Compliance Testing</title><link>https://bayesianhimanshu.org/blog/professional/verdict-validation-and-explainability-framework/</link><guid isPermaLink="true">https://bayesianhimanshu.org/blog/professional/verdict-validation-and-explainability-framework/</guid><description>Can You Trust an LLM to Grade Pharmaceutical Compliance? I Built a Framework to Find Out.</description><pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate><category>professional</category><category>evaluation</category><category>agents</category><category>llm-as-judge</category><category>gxp</category></item></channel></rss>