Why RAG

The LLM doesn't know PawWalk

In module 28 you saw what an LLM does: predict the next token from patterns it learned during training. That training happened long before this repo existed. So if you ask a raw model “What's PawWalk's cancellation policy?”, it has never seen docs/API-CONTRACT.md — it will confidently invent an answer. That made-up answer is called a hallucination, and it's the single biggest risk in shipping an LLM feature.

The model isn't lying on purpose. It's doing exactly its job — producing plausible text — with no access to your facts. The fix is to hand it the facts at question time.