OmniRAG - the right way to do Retrieval Augmented Generation with Azure Cosmos DB
summary
OmniRAG claims to pick the right source per query via intent detection and NL2Query, reducing LLM calls and cost; CosmosAIGraph implements it with OmniRAG logic, Azure Cosmos DB for persistence and built‑in vector search, an in‑memory indexed RDF triple store via Apache Jena, and Azure OpenAI (GPT‑4, text‑embedding‑ada‑002). Demo shows graph queries returning in ~5 ms, some answers served with zero tokens (direct DB/graph reads), session feedback for tuning intent routing, and utilities to build RDF from JSON.
AI-written summary · about
chapters (15)
- 0:00OmniRAG with Cosmos DB
- 1:08Basic RAG problems
- 5:03OmniRAG pattern benefits
- 7:58Knowledge graphs improve RAG
- 9:04OmniRAG design tenants
- 10:52Basic RAG vs OmniRAG
- 11:55Generating knowledge graphs
- 13:01Cosmos AI Graph demo
- 14:49Demo walkthrough
- 17:22Exploring Cosmos DB
- 18:48Sparkle query generation
- 19:51Direct DB queries
- 20:32Graph query execution
- 21:15Vector and graph search
- 22:36Summary and call to action
related talks
50mHow to design and build AI applications with vector search using Azure OpenAI & Azure Cosmos DB
11mScalable RAG with Azure Cosmos DB and DiskANN
21mHow RAG is Redefining AI Capabilities
12mBuilding Generative AI Apps with Azure Cosmos DB
03mMake RAGs easier with DataStax GitHub Copilot and Azure OpenAI
19mBeyond Retrieval: Adding a Reasoning Layer to RAG
30mGraph RAG: Bringing together graph and vector search to empower retrieval
39mEnhancing RAG with Neo4j Knowledge Graph