Day 4 — Advanced RAG Patterns

Basic RAG retrieves chunks and stuffs them into a prompt. Advanced RAG uses the LLM to decide what to retrieve, retrieves in multiple hops, handles structured data alongside text, and uses the agent itself to route queries. Today covers the patterns that move RAG from "works on a demo" to "works on real questions."

Lessons

1. Self-Querying Retrieval — the LLM generates structured filters from natural language queries
2. Multi-Hop Retrieval — retrieving information that requires connecting facts across multiple documents
3. Agentic RAG — the agent decides when and what to retrieve, not a fixed pipeline
4. Structured Data + Text — combining SQL queries with vector search for hybrid data
5. Conversational RAG — maintaining retrieval context across turns in a conversation

Next: Day 5 — Production RAG Systems