Phase 3: Agents and Memory — LLMs That Think and RememberPhase 1 was calling an LLM. Phase 2 was giving it documents to read. Phase 3 is giving it the ability to act. This post covers two things built this week: a multi-step agent that plans and executes usAug 14 · 10 min readlearn.sannan.app
From Zero to AI Engineer — 8 Levels That Actually Make SenseEvery AI engineering tutorial I found assumed too much. They started with LangChain, vector databases, and multi-agent orchestration before explaining what an LLM call actually looks like. I spent weeJul 22 · 12 min readlearn.sannan.app
Contract Intelligence — RAG over Legal DocumentsPhase 2 Week 1 built the foundation — embeddings, ChromaDB, a working RAG pipeline, and 6/6 evals. This week applied everything to a real domain: contract intelligence. Legal documents are the perfectJul 14 · 11 min readlearn.sannan.app
Embeddings and RAG — How AI Systems Actually Know ThingsPhase 1 ended with a working AI API. Eleven endpoints. Streaming chat. Structured output. Evals passing 15/15. Observable via Langfuse. But there was a problem I did not fully appreciate until Phase 21Jul 6 · 10 min readlearn.sannan.app
8 Weeks as an AI Engineer — What I Built, What Broke, What's Next Eight weeks ago I opened my first Python file and thought: how different can it be? I had four years of Angular. TypeScript every day. I knew how to build things people actually use. What I did not knJun 29 · 8 min readlearn.sannan.app
