<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Kaushal Prajapati</title><description>Kaushal Prajapati writes about production AI engineering, agentic systems, software engineering, deep-learning papers, evals, retrieval, and LLM platforms.</description><link>https://kforcode.dev/</link><language>en-us</language><item><title>The system layer behind AI products: what an agent runtime actually does</title><link>https://kforcode.dev/blog/the-system-layer-behind-ai-products/</link><guid isPermaLink="true">https://kforcode.dev/blog/the-system-layer-behind-ai-products/</guid><description>AI demos are easy; AI products are hard. The gap is a runtime — the shared execution layer that handles model routing, tool authorization, long-context state, memory, and fallback. Here is what it takes to run one at scale.</description><pubDate>Mon, 20 Jul 2026 00:00:00 GMT</pubDate><category>agents</category><category>ai-platform</category><category>runtime</category><category>llm</category><category>production</category></item><item><title>From 86% to 97%: engineering evidence retrieval for compliance agents</title><link>https://kforcode.dev/blog/evidence-retrieval-86-to-97/</link><guid isPermaLink="true">https://kforcode.dev/blog/evidence-retrieval-86-to-97/</guid><description>Compliance answers are only trustworthy if every claim cites the right evidence. Here is how hybrid retrieval, reranking, control-aware chunking, and citation-span ranking took evidence F1@10 from 86% to 97%.</description><pubDate>Tue, 14 Jul 2026 00:00:00 GMT</pubDate><category>rag</category><category>retrieval</category><category>evals</category><category>compliance</category><category>llm</category></item><item><title>One model instead of 23: consolidating a Document AI stack</title><link>https://kforcode.dev/blog/one-model-instead-of-23/</link><guid isPermaLink="true">https://kforcode.dev/blog/one-model-instead-of-23/</guid><description>A model per document type does not scale. Here is how a single layout-aware extraction architecture replaced 23+ document-specific models, cut OCR spend 70%, and made new document types cheap to add.</description><pubDate>Tue, 07 Jul 2026 00:00:00 GMT</pubDate><category>document-ai</category><category>ocr</category><category>ml-systems</category><category>cost</category><category>architecture</category></item></channel></rss>