<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Himanshu Patel — AI/ML Engineer</title><description>Writing on AI/ML engineering, LLM fine-tuning, RAG systems, and production machine learning.</description><link>https://himanshup.dev/</link><item><title>I Built a Local RAG System for Telecom Operations — Here&apos;s Everything I Learned</title><link>https://himanshup.dev/blog/telecomops-rag-local-llm/</link><guid isPermaLink="true">https://himanshup.dev/blog/telecomops-rag-local-llm/</guid><description>A full technical deep-dive into building TelecomOps RAG: two-machine LAN setup, 31K semantic chunks, Mistral 7B on an RTX 3050, and the real production gotchas nobody writes about.</description><pubDate>Sun, 31 May 2026 00:00:00 GMT</pubDate></item><item><title>Building Production RAG Systems for Telecom NOC</title><link>https://himanshup.dev/blog/rag-for-telecom-noc/</link><guid isPermaLink="true">https://himanshup.dev/blog/rag-for-telecom-noc/</guid><description>How we built TelecomOps RAG — a retrieval-augmented generation system for telecom network operations centers using Mistral 7B, ChromaDB, and FastAPI. Lessons from 31K chunks in production.</description><pubDate>Sat, 15 Nov 2025 00:00:00 GMT</pubDate></item><item><title>Fine-tuning LLMs with QLoRA: A Practical Production Guide</title><link>https://himanshup.dev/blog/qlora-finetuning-guide/</link><guid isPermaLink="true">https://himanshup.dev/blog/qlora-finetuning-guide/</guid><description>A complete walkthrough of QLoRA fine-tuning — from dataset preparation to production inference. What the papers don&apos;t tell you, and what actually breaks in practice.</description><pubDate>Wed, 22 Oct 2025 00:00:00 GMT</pubDate></item><item><title>Agentic AI Patterns That Actually Work in Production</title><link>https://himanshup.dev/blog/agentic-ai-production-patterns/</link><guid isPermaLink="true">https://himanshup.dev/blog/agentic-ai-production-patterns/</guid><description>Most agentic AI demos are impressive. Most agentic AI production systems are fragile. Here are the patterns that survive contact with real users, real APIs, and real failure modes.</description><pubDate>Wed, 10 Sep 2025 00:00:00 GMT</pubDate></item></channel></rss>