ModelRefs / Advanced RAG Stack — Architecture Pattern
Advanced RAG Stack — Architecture Pattern
Hybrid retrieval, query rewriting, reranking, evaluation harness, and source-attributed generation in a microservice topology.
Overview
Production-grade RAG: hybrid lexical+vector retrieval, query rewriting, cross-encoder reranking, answer evaluation, and continuous offline evals. Each layer is independently scalable.
When to use it: You need high-accuracy RAG at scale with eval-grade observability.
Pattern details
| Pattern class | rag |
|---|---|
| Difficulty | advanced |
| Topology | microservices |
| Also known as | production rag, hybrid rag |
| Last reviewed | 2026-06-07 |
Known failure modes
- Reranker bottleneck — Cross-encoder dominates latency. Mitigation: Cache; reduce N; batch.
- Eval drift — Offline evals drift from production. Mitigation: Anchor evals to production traces.
When not to use it
- Skipping eval harness in production RAG.
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Advanced RAG Stack — Architecture Pattern.
Frequently asked questions
When should I adopt the Advanced RAG Stack?
You need high-accuracy RAG at scale with eval-grade observability.
What are common failure modes of Advanced RAG Stack?
Reranker bottleneck • Eval drift
Is Advanced RAG Stack production-ready?
Yes when paired with the safety controls and observability hooks documented on the pattern page.