Production RAG Systems

Production RAG Systems is a practical engineering guide to building retrieval-augmented generation that works outside the notebook. It starts from a simple observation: the quality of a RAG system is bounded by the quality of its retrieval, and the quality of its retrieval is bounded by decisions most teams make in an afternoon and regret for a quarter.

The book walks through the full production stack — embeddings and vector representations, HNSW and IVF index structures, chunking strategies that respect document structure, hybrid search that combines dense and sparse retrieval with Reciprocal Rank Fusion, cross-encoder re-ranking, query understanding and rewriting, and context assembly that keeps generation grounded.

It covers the failure modes that destroy production systems: a retrieval miss that produces a confident hallucination, an embedding upgrade that silently invalidates every cached vector, a permission filter applied after retrieval that leaks documents across tenants, a chunking change that shifts recall by twenty points on a specific query class. Each is presented with the failure, the countermeasure, and the operational tradeoff.

Fourteen chapters. Real Python code. No toy examples. Written for engineers who need RAG to work at scale, not in a demo.