Production RAG at Scale with Azure Database for PostgreSQL
summary
Serenity Star runs RAG on Azure Database for PostgreSQL using pgvector for embeddings, exploring DiskANN for approximate nearest-neighbor search, and integrating with Microsoft Semantic Kernel; multiple embedding models force different vector sizes so the team creates per-model/per-size tables. Multi-tenant versioning produces growth to 1.4M records, so a multi-column index (tenant, agent, version) drops similarity-query latency from ~750ms average to ~50–70ms and eliminates expensive sequential scans under high concurrency.
AI-written summary · about
chapters (12)
- 0:00Music & introduction
- 0:47About us: Serenity Star and AI agents
- 2:41The four dimensions of an AI agent
- 4:08Demo: from knowledge upload to agent answer
- 8:07The four knowledge processing stages
- 9:02Agent execution pipeline: from query to LLM
- 10:08Architecture overview and the Postgres choice
- 12:43Multiple embedding models, different vector sizes
- 14:16Multi-tenant applications, versioning at 1.4M records
- 15:23Similarity search and a multi-column index fix
- 20:03Exploring DiskANN for vector search
- 21:25Benefits in production: Azure Database for PostgreSQL scalability and monitoring
related talks
20mBuilding Enterprise RAG with Azure Database for PostgreSQL and pgvector
12mThe mechanics of integrating AI and vector search into Azure Database for PostgreSQL
44mPostgres and OpenAI - a match made in Azure
21mBuilding Intelligent Applications with Graph-Based RAG on PostgreSQL
50mUnleashing the Power of Azure for PostgreSQL
47mRAG at scale: production-ready GenAI apps with Azure AI Search
47mExtendable by Design: Building Generative AI Apps with Postgres and Vector Storage and Azure AI
27m6 things you can do with azure_ai & PostgreSQL on Azure