Advanced RAG Patterns — Part 1: Hybrid Retrieval, Query Craft & Agentic RAG — 43 Templates & Checklists for Hybrid Retrieval & Rank Fusion, Query Transformation & Routing, Agentic Loops, Self-Correction (CRAG & Self-RAG), and Multi-Hop Retrieval

Your RAG system demos beautifully and plateaus in production — this is the working half of the discipline that breaks the plateau. 43 production templates & checklists distilled from Chapters 1–6 of Advanced RAG Patterns: decompose the accuracy ceiling before treating it, engineer hybrid retrieval and rank fusion properly, fix the query before it reaches an index, run agentic retrieval inside hard budgets, grade evidence before generation trusts it (CRAG & Self-RAG), and follow chained questions hop by hop in an evidence ledger. PDF + Markdown source.

The demo scored 91%. In the room, on a real question, it scored zero.

The question was innocent — whether a flood endorsement covers detached garages in Zone B for a dwelling built before 2011. The system returned a confident, well-written, beautifully cited answer. It was also wrong: the top passages covered flood cover for main dwellings, and none mentioned garages or the pre-2011 exception. That is the plateau — the stopping point where single-pass RAG, one retrieval decision made with the user's raw words against one index, stops improving no matter how you tune it. Advanced RAG Patterns — Part 1 is the working discipline for breaking it: 43 copy-paste templates and checklists distilled from Chapters 1–6 of the advanced patterns book — name the failure shape and decompose the ceiling first, give the corpus many doors and fuse their rankings defensibly, fix the query before it reaches an index, let a bounded agent loop decide the next step, grade the evidence before generation trusts it, and follow chained questions one ledgered hop at a time.

Every template carries the book's tested positions built in: failures counted in buckets before any cure is chosen, the oracle test that proves whether the model is really the bottleneck, RRF computed by hand on the book's own worked table (document A wins at 0.04840 on consensus, not dominance), BM25 engineered analyzer-first before it is replaced, structured retrieval treated as a first-class door, a control loop whose examiner — not its planner — decides when evidence is enough, budgets counted in code and never in prompts, CRAG's three verdicts each bound to a real action, and hop budgets that stop at three because recall flattens while noise compounds. Where the book gives a number — its fusion constant, its budget starting points, its illustrative outcome figures — the template carries the book's number and labels it as the book's, illustrative where the book says illustrative. Bracketed fields are yours to fill; nothing here invents a threshold for your system.

What's inside — 43 templates & checklists in 7 sections

1. The Plateau — Why Single-Pass RAG Stops Improving (6) — Four Plateau Symptoms Triage Sheet · Accuracy-Ceiling Decomposition Worksheet · Oracle Test Protocol · Usual-Knobs Audit · Four Control Layers Adoption Map · Complexity Budget & Stay-Simple Decision Sheet

2. Hybrid Retrieval & Rank Fusion, Engineered (7) — Complementarity & Oracle-Recall Overlap Plot · BM25 Engineering Worksheet · Learned Sparse & Structured Retrieval Decision Sheet · Lane Sizing & Normalization Decision Record · Reciprocal Rank Fusion Worked Sheet · Weighted & Learned Fusion Control Sheet · Fusion Measurement & Partial-Failure Policy

3. Query Transformation, Decomposition & Routing (7) — Four Query-Failure Patterns Diagnostic · Conversational Rewriting Contract · HyDE Protocol · Step-Back & Multi-Query Fan Sheet · Decomposition Plan Sheet · Routing Decision Table & Fallback Specification · Per-Transform Measurement & Intent-Drift Guardrail Sheet

4. Agentic RAG & the Control Loop (6) — Question Taxonomy Routing Sheet · The Control Loop Protocol · Five-Part Anatomy & Tool Registry Specification · Evidence Ledger & Sufficiency Criteria Sheet · Budget Envelope Configuration · Agent Failure-Mode Silhouettes & the Honest Pipeline Case

5. Self-Correcting Pipelines — CRAG & Self-RAG (6) — Evidence-Failure Classification · CRAG Three-Verdict Controller Sheet · Grader Family Selection & Cascade Design · Knowledge Refinement & Web Fallback Sheet · Self-RAG Reflection Sheet · Threshold Policy, Correction Cost Ledger & the Kill Switch

6. Multi-Hop Retrieval for Questions That Chain (6) — Chained-Question Classification Sheet · Plan-First Chain Specification · IRCoT Interleaving Protocol & Style-Choice Sheet · Evidence Ledger, State Discipline & Hop Budget Sheet · Per-Hop Evaluation Scorecard · Hop Drift Defence & Graph-Boundary Router

7. Part 1 Gates & Working Artefacts (5) — Failure-Triage & Baseline-Freeze Artefact · Fusion Measurement & Lane-Failure Gate · Query-Layer Build-Order Gate · Multi-Hop Chain Audit Sheet · Part 1 Advanced Retrieval Gate — Chapters 1–6 Sign-Off — the book's six chapter Key Takeaways consolidated into one evidence-backed gate (36 lines)

Who this is for

RAG and platform engineers whose baseline system has stopped improving, search and retrieval leads who own hybrid stacks in production, and AI architects deciding — with evidence, not fashion — which advanced pattern a measured symptom actually justifies. The source book assumes you already run the baseline pipeline (ingestion, chunking, embeddings, a vector index, grounded generation); this pack begins exactly where that baseline stops improving. If your team is still arguing about whether the model is the problem, start at Template 3.

What this pack is not

Not Part 2. Context engineering and assembly, reranking in depth, evaluation as a full discipline, the failure-mode catalogue in depth, adaptive retrieval control, and the assembled reference platform are the book's Chapters 7–12, with Appendices A–B. They live in the Part 2 companion pack (Product #26); nothing here restates them, and this pack is complete and useful without them.

Not the baseline. Ingestion, chunking, embeddings, vector-index operations, and grounded generation are the RAG Architecture book's ground. Product #2 (RAG Architecture Templates) covers that baseline pipeline; this pack assumes it, keeps it as the frozen baseline every pattern must beat, and goes beyond it — it does not re-teach it.

No runnable production code or tooling. The book's Python listings (the triage harness, the fusion functions, the query-understanding pipeline, the agentic loop, the CRAG controller, the multi-hop retriever) are standard-library teaching implementations, described in the book as skeletons of the production patterns. The pack ships the templates that specify, configure, audit, and operate these patterns — not a production retrieval platform.

No benchmark, accuracy, or results claims. Every outcome figure in the source chapters is the book's illustrative figure, from fictional worked stories, and the templates label it exactly that way. Nothing here promises that any pattern will lift your system's accuracy by any amount — each template instead ships the measurement (triage counts, oracle test, per-lane/fused/oracle recall, golden-set gain) by which you prove it on your own corpus.

No guarantee of correctness. The book's own position: advanced patterns are adopted against measured symptoms, inside budgets, or not at all — and an honest abstention is a successful outcome. These templates manage, measure, and bound retrieval behaviour; they never promise that an answer is right.

Companion products

Advanced RAG Patterns — Part 2 (Product #26) is this pack's sibling, built from Chapters 7–12 and Appendices A–B of the same book — context engineering, reranking, evaluation, and the platform: a companion for when you are ready for the second half, never a requirement for this pack to work. Product #2, RAG Architecture Templates, covers the baseline pipeline this pack assumes. This pack stands alone.

Note: the $99 Architecture Templates Bundle currently contains Products #1–#3 only. This pack launches standalone; no bundle inclusion is promised for this product.

Built by the author of Advanced RAG Patterns: Hybrid, Agentic, and Self-Correcting Retrieval Architectures, from the retrieval discipline behind production enterprise AI systems.

About the author

Naveen Kumar Josyula is the author of 35 AI books on Amazon KDP and the CEO and founder of AISoltrics Technologies Inc. He is the author of the source book, Advanced RAG Patterns: Hybrid, Agentic, and Self-Correcting Retrieval Architectures, from which every template in this pack is distilled.

Support promise

If anything in this file is wrong or unclear, email support@aisoltrics.com before you judge it — we fix verified issues within one business day and send you the corrected file free. (You can also reach the founder's office at aadhya@aisoltrics.com.) Updates to this product are free to past buyers via your library. Sold by AISoltrics Technologies Inc.