Basic vector search is a good starting point, but most real RAG systems need more than that.

A user asks, "What is the refund policy?" The relevant document says, "Customers may return physical products within 30 days of delivery." The meaning is close, but the words do not match. Another user asks about "SSO setup," while the docs say "SAML identity provider configuration." A third user searches for an error code, where an exact match matters more than semantic similarity.

These are retrieval problems, not generation problems. If the right evidence never reaches the model, a better prompt cannot reliably fix the answer.

This chapter covers the retrieval techniques that usually become useful after the first RAG prototype: query expansion, HyDE, hybrid search, reranking, parent-child retrieval, contextual retrieval, multi-step retrieval, and structured-data routing.