A RAG node in the graph

The last mile

In Module 30 you built a RAG pipeline from scratch in the playground: chunk the docs, embed with bag-of-words, cosine-rank, retrieve the top chunks, and stuff them into the prompt. It answered "what's PawWalk's cancellation policy?" by quoting the real doc instead of hallucinating.

That retriever is still sitting in playground/rag-pawwalk/. This module moves it into the real assistant — the LangGraph state machine from Module 29 — so an owner can ask a policy question in the same chat where they book walks. Then you'll measure whether it actually answers well, and graduate.

Nothing here is new. It's Module 29's graph plus Module 30's retriever, wired together with the conditional-edge trick you already know.