rag-architectures: Curated Catalog of Production RAG Patterns and Implementations

RAG is the workhorse of 2026 AI applications. Every serious AI product - customer support, code assistants, legal review, medical Q&A - has some RAG component. But the term covers a huge range of patterns: simple vector search, hybrid search, re-ranking, query rewriting, agentic RAG, graph RAG, multi-modal RAG. Picking the right one for your use case is non-trivial.

rag-architectures collects the patterns that actually work in production, with reference implementations you can study or adapt. Each pattern comes with: the architecture diagram, the code, the trade-offs, and links to the original research.

What's in the catalog

Naive RAG (just vector search). Advanced RAG (query rewriting, re-ranking, hybrid search). Modular RAG (specialized modules you mix and match). Agentic RAG (the LLM decides which retrieval strategy to use). Graph RAG (entities and relationships, not just chunks). Self-RAG (the model critiques its own retrieval). Each with working code.

Why this is useful

Most 'RAG tutorials' online show the naive version and stop there. This repo goes all the way to agentic and graph RAG, which is where the real production value is. If you're serious about RAG, you need to understand these patterns - and the code in this repo is the fastest way to get there.

Verdict

The most useful RAG reference repo I've seen in 2026. Treat it as required reading before you build any non-trivial RAG system. The maintainer clearly uses these patterns in production, and the code quality reflects that.