Skip to content
RAG Field Guide for AI Agents
The Trove library

RAG Field Guide for AI Agents

Load when designing, building, debugging, evaluating or securing a RAG system: choosing RAG vs fine-tuning vs long context, parsing PDFs, chunking, embeddings, vector stores and HNSW, hybrid search, reranking, query rewriting, grounding, evaluation and LLM judges, caching, monitoring.

Read a preview

You can read the contents and the first 2,000 characters.
Pro includes the whole book.

  • By Sahar Carmel
  • 37 chapters
  • Updated Nov 11, 2026
  • 7 readers

What’s inside

  • RAG or not: choosing the adaptation path
  • Pipeline architecture and failure map
  • Ingestion, parsing and index sync
  • Chunking
  • Embeddings
  • Vector stores and ANN indexing

What is RAG Field Guide for AI Agents?

Load when designing, building, debugging, evaluating or securing a RAG system: choosing RAG vs fine-tuning vs long context, parsing PDFs, chunking, embeddings, vector stores and HNSW, hybrid search, reranking, query rewriting, grounding, evaluation and LLM judges, caching, monitoring, poisoning and prompt-injection defenses. Every rule is a RULE/WHY/APPLY entry citing a page in one of five source books (Kimothi, Huyen, Grainger/Turnbull/Irwin, Iusztin/Labonne, Rothman) or one of 23 open-access research papers (arXiv).

Read more

markdown · Listed Nov 10, 2026

  • rag
  • retrieval
  • llm
  • ai-agents
  • evaluation

From the first page

What this is. A distilled, agent-readable synthesis of five RAG and retrieval books and 23 open-access research papers. It is an original paraphrase with page citations, not a copy of any chapter; quotations are limited to a few words. Every rule points to the page where the source states or demonstrates it, so you can verify before acting.

Audience and governance. Public reference for engineers and their AI agents. Custodian: the listing author. Licence: all rights reserved by the author; the five source books remain the property of their authors and publishers, and this guide is an independent summary that cites them. Attribution: see the source key below. Last updated: 2026-11-10. Scope of validity: five books from 2024 to 2025 (one is an unfinished early-access edition, see Ch 26) plus 23 open-access papers from 2016 to 2026 (Ch 19 to Ch 25). Next review: 2027-05-10 (every 6 months), because model names, prices and benchmark leaders age fast; principles age slowly.

Page numbers refer to the ebook edition of each book that the author read, which can differ from print pagination (for AI-Powered Search, printed page = cited page minus 5). Use them to locate a passage; expect small differences in other editions.

Preview · 2,000 of 361,302 characters

Data & Database Review for AI Agents
Awesome Design.md — A Curated Collection of Design Resources in Markdown
Dynamic Workflows in Claude Code: A Complete Guide to Multi-Agent Harnesses
SEO for AI Agents

Contents

  1. 01RAG Field Guide for AI Agentsp.1
    1. Source keyp.1
    2. How to use this bookp.1
  2. 02Decision Matrixp.2
  3. 03RAG or not: choosing the adaptation pathp.3
    1. Quick referencep.3
    2. Agent rulesp.3
    3. Additional rules (third pass)p.3
  4. 04Pipeline architecture and failure mapp.4
    1. Quick reference: symptom to stage to fixp.4
    2. Agent rulesp.4
    3. Additional rules (third pass)p.4
  5. 05Ingestion, parsing and index syncp.5
    1. Quick referencep.5
    2. Agent rulesp.5
    3. Additional rules (third pass)p.5
  6. 06Chunkingp.6
    1. Quick reference: every number the sources give (all are demos, none is tuned)p.6
    2. Agent rulesp.6
    3. Additional rules (third pass)p.6
  7. 07Embeddingsp.7
    1. Quick referencep.7
    2. Agent rulesp.7
    3. Additional rules (third pass)p.7
  8. 08Vector stores and ANN indexingp.8
    1. Quick reference: ANN algorithms (from AE p.225, LEH p.187-188, APS p.340-341)p.8
    2. Agent rulesp.8
    3. Additional rules (third pass)p.8
See all 37 parts
  1. 09Retrieval: lexical, dense, hybrid, filters, routingp.9
    1. Quick referencep.9
    2. Agent rulesp.9
    3. Additional rules (third pass)p.9
  2. 10Query understanding and transformationp.10
    1. Quick referencep.10
    2. Agent rulesp.10
    3. Additional rules (third pass)p.10
  3. 11Reranking and result shapingp.11
    1. Quick referencep.11
    2. Agent rulesp.11
    3. Additional rules (third pass)p.11
  4. 12Context assembly and generationp.12
    1. Quick referencep.12
    2. Agent rulesp.12
    3. Additional rules (third pass)p.12
  5. 13Advanced patternsp.13
    1. Quick reference: which pattern for which failurep.13
    2. Agent rulesp.13
    3. Additional rules (third pass)p.13
  6. 14Evaluationp.14
    1. Quick reference: metricsp.14
    2. Agent rulesp.14
  7. 15Chapter 14 (continued 1) — Evaluationp.15
    1. Additional rules (third pass)p.15
  8. 16Chapter 14 (continued 2) — Evaluationp.16
    1. Additional rules (third pass)p.16
  9. 17Relevance engineering with signalsp.17
    1. Agent rulesp.17
    2. Additional rules (third pass)p.17
  10. 18Production: latency, cost, caching, monitoring, driftp.18
    1. Quick referencep.18
    2. Agent rulesp.18
  11. 19Chapter 16 (continued) — Production: latency, cost, caching, monitoring, driftp.19
    1. Additional rules (third pass)p.19
  12. 20Security, privacy and guardrailsp.20
    1. Quick referencep.20
    2. Agent rulesp.20
    3. Additional rules (third pass)p.20
  13. 21Anti-patterns and red flagsp.21
  14. 22Document parsing and OCR (primary sources)p.22
    1. Quick referencep.22
    2. Agent rulesp.22
    3. Caveatsp.22
  15. 23Chapter 19 (continued 1) — Document parsing and OCR (primary sources)p.23
    1. Additional rules (third pass)p.23
  16. 24Chapter 19 (continued 2) — Document parsing and OCR (primary sources)p.24
    1. Additional rules (third pass)p.24
  17. 25Retrieval techniques from the primary papersp.25
    1. Quick referencep.25
    2. Agent rulesp.25
    3. Additional rules (third pass)p.25
    4. Caveatsp.25
  18. 26Structure-aware indexing: late chunking, RAPTOR and graph RAGp.26
    1. Quick referencep.26
    2. Agent rulesp.26
    3. Caveatsp.26
  19. 27Chapter 21 (continued 1) — Structure-aware indexing: late chunking, RAPTOR and graph RAGp.27
    1. Additional rules (third pass)p.27
  20. 28Chapter 21 (continued 2) — Structure-aware indexing: late chunking, RAPTOR and graph RAGp.28
    1. Additional rules (third pass)p.28
  21. 29Self-reflective and agentic RAGp.29
    1. Quick referencep.29
    2. Agent rulesp.29
    3. Additional rules (third pass)p.29
    4. Caveats and correctionsp.29
  22. 30Context position, long context, embedders and rerankersp.30
    1. Quick referencep.30
    2. Agent rulesp.30
    3. Caveatsp.30
  23. 31Chapter 23 (continued 1) — Context position, long context, embedders and rerankersp.31
    1. Additional rules (third pass)p.31
  24. 32Chapter 23 (continued 2) — Context position, long context, embedders and rerankersp.32
    1. Additional rules (third pass)p.32
  25. 33Evaluation: definitions and the research surveyp.33
    1. Quick referencep.33
    2. Agent rulesp.33
    3. Additional rules (third pass)p.33
    4. Caveatsp.33
  26. 34Security: poisoning, injection and defenses (primary sources)p.34
    1. Quick referencep.34
    2. Agent rulesp.34
    3. Caveatsp.34
  27. 35Chapter 25 (continued) — Security: poisoning, injection and defenses (primary sources)p.35
    1. Additional rules (third pass)p.35
  28. 36Source map, source quality and known gapsp.36
    1. Which book for whatp.36
    2. Third pass: previously skipped pages (2026-11-10)p.36
    3. Primary papers added 2026-11-10p.36
    4. Known gaps after the second passp.36
  29. 37Self-audit and gap mapp.37
    1. Knowledge-asset checksp.37
    2. Agent-readability checksp.37
    3. Corrections made in the second passp.37
    4. Corrections made in the third passp.37
    5. Verification logp.37
    6. Open gaps (dated 2026-11-10)p.37