Extendable by Design: Building Generative AI Apps with Postgres and Vector Storage and Azure AI
summary
Shows a working Rails app that loads ~30 years of mailing‑list emails (1998–Mar 2024), adds a generated ts_vector column for full‑text search, creates 1536‑dim embeddings via Azure OpenAI, stores them with pgvector, and compares IVF vs HNSW indexes. Demonstration runs vector search and RAG: initial sequential scans of ~10+ ms drop to <1 ms with an HNSW index; summarization and grounded answers come from retrieved emails, with occasional prompt‑level failures during the live demo.
AI-written summary · about
chapters (22)
- 0:00Postgres and generative AI
- 1:10Postgres extendable by design
- 3:00Postgres extensions for features
- 7:37Mailing list search helper
- 8:39Email search challenges
- 10:03Full text search in Postgres
- 13:15How Postgres tokenizes text
- 15:03Highlighting search matches
- 16:06Ranking full text queries
- 18:36Vector storage with PG Vector
- 19:22Vector indexing: IVF vs HNSW
- 22:11Azure AI for embeddings
- 23:38Retrieval augmented generation
- 26:27Email embeddings generation
- 28:13Semantic email retrieval
- 29:18Optimizing vector queries
- 30:23Testing RAG with Azure AI Studio
- 31:50Live demo of RAG
- 34:43Refining prompts for responses
- 37:15Hybrid search: text and vectors
- 38:42Exploring email dataset
- 42:55Q&A
related talks
12mThe mechanics of integrating AI and vector search into Azure Database for PostgreSQL
22mFrom Postgres full text search to Retrieval Augmented Generative search
20mBuilding Enterprise RAG with Azure Database for PostgreSQL and pgvector
40mUnleashing the Power of Vector Search with Azure SQL Database
27m6 things you can do with azure_ai & PostgreSQL on Azure
01h 16mAzure SQL ❤️ Open AI
44mPostgres and OpenAI - a match made in Azure
33mVector Databases for GenAI applications