RAG-08 · Hybrid search and re-ranking¶
Teams ship with default retrieval settings and tune weights and re-rankers by guesswork, so a demo that impressed fails on real questions. When relevance drops in production nobody can tell whether retrieval, ranking or the prompt caused it.
- Who reads it: GenAI engineers and RAG leads tuning retrieval, and DevOps engineers who own search configuration in production
- Outcome: Measured relevance before every ranking change
- Pack owner: Deeparani (catalog owner role: RAG Lead)
Master catalog entry¶
- Catalog name: Hybrid Search & Re-ranking Playbook (P2)
- Why it matters: The difference between a demo that impresses and one that disappoints
- What it removes: Tuning recipe instead of guesswork on weights and re-rankers
- Stack: Azure AI Search semantic ranker, Cohere rerank, RRF
- Maturity target: Level 2 — benchmarked on three corpus types
Playbooks¶
| ID | Story | Title | Kind | Complexity | Urgency | Status | Owner | Planned publish |
|---|---|---|---|---|---|---|---|---|
| PB-RAG-08-01 | RAG-08-S1 | Choose the retrieval mode: keyword, vector, hybrid, semantic ranker or external re-ranker | Decision guide | Medium | 3 | draft | Deeparani | 2026-12-04 |
| PB-RAG-08-02 | RAG-08-S2 | Tune hybrid weights and re-ranking against a golden query set | How-to | Medium | 3 | draft | Gayathri Ramachandran | 2026-12-16 |
| PB-RAG-08-03 | RAG-08-S3 | Diagnose and fix a relevance regression in production | Runbook | High | 5 | draft | Deeparani | 2026-11-16 |
| PB-RAG-08-04 | RAG-08-S4 | Relevance go-live and ranking change checklist | Checklist | Low | 4 | draft | Gayathri Ramachandran | 2026-11-20 |