Building an Enterprise Knowledge Management System using Hybrid Search in Azure Cosmos DB
summary
A working NestJS demo shows an enterprise knowledge-management API using Azure Cosmos DB NoSQL with JavaScript/TypeScript SDK, Azure OpenAI embeddings (deployment text-embedding-3-large) and vector + BM25 full‑text indexes fused by Reciprocal Rank Fusion. Indexes use 432‑dim vectors and Cosine distance; the app auto‑creates the DB, generates 32 fake items, and serves vector, full‑text and hybrid endpoints (top‑10 default). Hybrid fusion improves ranking in demo but brittle results appear with heavy typos.
AI-written summary · about
chapters (27)
- 0:00Enterprise knowledge systems
- 1:07Hybrid search and benefits
- 2:13Vector search in Cosmos DB
- 3:37Full text search fundamentals
- 5:26Reciprocal rank fusion
- 6:33Azure OpenAI and Cosmos DB setup
- 10:38NestJS framework overview
- 13:06Cosmos DB SDK dependencies
- 14:10Database service structure
- 17:21Knowledge item entity design
- 18:24Hybrid search indexing policy
- 19:27Vector embedding policy setup
- 20:09Full text search language config
- 20:50Querying without search
- 21:35Vectorizing queries
- 22:59Full text search queries
- 23:41Hybrid search implementation
- 25:07Search endpoints controller
- 25:54Sample data generation
- 27:52Application setup verification
- 28:34Vector embeddings generation
- 31:22Normal query execution
- 31:45Vector search with nearest neighbor
- 33:30Full text search execution
- 34:54Hybrid search demonstration
- 36:41Search methods limitations
- 37:01Summary
related talks
50mHow to design and build AI applications with vector search using Azure OpenAI & Azure Cosmos DB
12mBuilding Generative AI Apps with Azure Cosmos DB
20mBuilding AI Applications with Azure Cosmos DB JavaScript SDK
27mKeynote
03h 12mAzure Cosmos DB Conf 2024
40mUnleashing the Power of Vector Search with Azure SQL Database
28mCreating Enterprise Knowledge Graphs from Unstructured Data
01h 16mAzure SQL ❤️ Open AI