RAG-11 · GraphRAG pattern kit¶
Clients ask for entity and relationship reasoning, and teams start a costly graph build without proving that hybrid RAG cannot answer the questions. Graph indexing and refresh costs are rarely modelled, so pilots stall when the bill arrives.
- Who reads it: RAG leads and senior GenAI engineers assessing or building graph-based retrieval
- Outcome: Go or no-go on GraphRAG backed by cost and quality evidence
- Pack owner: Deeparani (catalog owner role: RAG Lead)
Master catalog entry¶
- Catalog name: GraphRAG Pattern Kit (P3)
- Why it matters: Differentiator for multi-hop and relationship-heavy questions
- What it removes: Avoids a speculative build when a client asks for entity reasoning
- Stack: Neo4j or Cosmos Gremlin, LangChain, Azure OpenAI
- Maturity target: Level 1 — experimental; cost model must be validated per case
Playbooks¶
| ID | Story | Title | Kind | Complexity | Urgency | Status | Owner | Planned publish |
|---|---|---|---|---|---|---|---|---|
| PB-RAG-11-01 | RAG-11-S1 | Decide whether GraphRAG is worth it | Decision guide | Medium | 2 | draft | Yogeshwaran | 2026-12-14 |
| PB-RAG-11-02 | RAG-11-S2 | Build a GraphRAG proof of concept with Azure OpenAI and Neo4j or Cosmos DB for Apache Gremlin | How-to | High | 2 | draft | Deeparani | 2026-12-14 |
| PB-RAG-11-03 | RAG-11-S3 | GraphRAG cost model and go-live checklist | Checklist | Medium | 3 | draft | Gayathri Ramachandran | 2026-12-23 |
| PB-RAG-11-04 | RAG-11-S4 | Refresh the knowledge graph after corpus, prompt or version changes | Runbook | High | 3 | draft | Deeparani | 2026-12-08 |