SASANAN ANSARIinlearn.sannan.app·5h ago · 10 min readEmbeddings 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 210
SASANAN ANSARIinlearn.sannan.app·Jun 29 · 8 min read8 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 kn00
SASANAN ANSARIinlearn.sannan.app·Jun 29 · 10 min readEvals and Observability — Knowing If Your AI Actually WorksWeek 6 ended with four structured endpoints and a feeling that things were working. The Darjeeling details looked right. The sentiment analysis felt accurate. But feelings are not engineering. This we00
SASANAN ANSARIinlearn.sannan.app·Jun 22 · 8 min readWeek 6:Structured Output — Getting the LLM to Return Reliable JSONWeek 5 ended with a working chat app — Angular frontend streaming responses from a FastAPI backend. The model answered in plain text. Useful for conversation. Useless the moment you want to build anyt00
SASANAN ANSARIinlearn.sannan.app·Jun 18 · 12 min readConnecting Angular to My AI Backend — Full Stack AI in PracticeWeek 4 ended with api.sannan.app live — a real public AI endpoint anyone could call. Week 5 is about building the thing people actually use to call it. This week I built the Angular frontend at ai-cha00
