Building Retrieval-Augmented Generation Systems That Actually Work

Read before you buy: the PDF here is a free sample — the first 20 pages of the 439-page book (payments, sales and reviews on this demo site are sample data). Get the full e-book at thenexbooks on Gumroad.

Your RAG demo worked on ten documents. Production has ten thousand — and they fight back.

Retrieval-augmented generation looks simple: embed some documents, search, paste the results into a prompt. Then real data arrives. PDFs with broken tables, policies that differ by region and change every quarter, part numbers that embeddings can't match, users who may only see some documents, questions the corpus can't answer, and instructions hidden inside the content itself. Most RAG systems fail quietly: they answer confidently from the wrong passage.

This book treats RAG as an engineering discipline: start with evaluation, understand every stage of the pipeline, measure each change, and design for the failure modes you will certainly meet. Vendor-neutral, with no invented benchmarks.

• 439 pages · 28 chapters + 4 appendices
• Build a complete pipeline end to end, then improve it stage by stage with a golden dataset
• Ingestion of PDFs, HTML, Office files, and tables; chunking strategies that preserve meaning
• Metadata, enrichment, vector databases (pgvector and dedicated engines), and keeping indexes fresh
• Dense, sparse, and hybrid search; query rewriting and decomposition; reranking and fusion
• Access control and multi-tenant retrieval; parent-child, hierarchical, and graph-based RAG
• Grounded prompting, citations, answer verification, and knowing when not to answer
• Evaluation with retrieval metrics and LLM-as-judge, agentic RAG, latency, cost, and caching
• Security (prompt injection, data leakage, PII), observability, and scaling
• One realistic running example (Kestrel Cycles, a bicycle and e-bike maker) and two full case studies

Table Of Contents

Part I — Foundations

• Ch 1 · Why RAG: What Retrieval Solves and What It Doesn't
• Ch 2 · LLMs, Tokens, and Context Windows for RAG Builders
• Ch 3 · Embeddings and Vector Similarity Explained
• Ch 4 · Your First RAG Pipeline, End to End
• Ch 5 · Defining Success: Requirements, Metrics, and a Golden Dataset

Part II — Ingestion and Indexing

• Ch 6 · Document Ingestion: PDFs, HTML, Office Files, and Tables
• Ch 7 · Chunking Strategies That Preserve Meaning
• Ch 8 · Metadata, Enrichment, and Document Structure
• Ch 9 · Vector Databases and Indexes: pgvector, Dedicated Engines, and Trade-offs
• Ch 10 · Keeping the Index Fresh: Updates, Deletes, and Versioning

Part III — Retrieval

• Ch 11 · Dense, Sparse, and Hybrid Search
• Ch 12 · Query Understanding: Rewriting, Expansion, and Decomposition
• Ch 13 · Reranking and Result Fusion
• Ch 14 · Filtering, Access Control, and Multi-Tenant Retrieval
• Ch 15 · Advanced Retrieval: Parent-Child, Hierarchical, and Graph-Based RAG

Part IV — Generation

• Ch 16 · Prompting for Grounded Answers
• Ch 17 · Citations, Attribution, and Answer Verification
• Ch 18 · Knowing When Not to Answer: Abstention and Fallbacks
• Ch 19 · Structured Outputs, Tables, and Multimodal RAG
• Ch 20 · Conversational RAG: Memory and Follow-Up Questions

Part V — Production

• Ch 21 · Evaluating RAG: Retrieval Metrics, Answer Quality, and LLM-as-Judge
• Ch 22 · Agentic RAG: Tools, Routing, and Multi-Step Retrieval
• Ch 23 · Latency, Cost, and Caching
• Ch 24 · Security and Privacy: Prompt Injection, Data Leakage, and PII
• Ch 25 · Observability, Feedback Loops, and Continuous Improvement
• Ch 26 · Deploying and Scaling RAG Services

Part VI — Case Studies

• Ch 27 · Case Study: Kestrel Assist — A Customer Support Assistant
• Ch 28 · Case Study: Spoke — Engineering Knowledge Search with Access Control

Appendices

• App A · RAG Cheat Sheet
• App B · Prompt and Configuration Templates
• App C · Production Checklist
• App D · Glossary

WHO IT IS FOR
Software, backend, and AI engineers building question answering or search over their organization's documents; data and platform engineers responsible for ingestion pipelines and vector infrastructure; and tech leads who need to judge whether a RAG system is actually good enough to ship.

FORMAT
PDF (A4), 439 pages, English, clickable table of contents and PDF bookmarks. 1st Edition, 2026. Published by thenexbooks.

Book details