From Transactions to Intelligence: Postgres for Open AI & Analyti
summary
Three concrete pathways: WarehousePG MPP forks Greenplum into a scale‑out Postgres that reaches petabyte class with built‑in ML and external table federation; extensions like pgvector vs GPU‑accelerated VectorCord show a 100M‑vector benchmark where GPU + VectorCord yields thousands QPS vs ~1k for pgvector and handles 400M vectors that HNSW breaks on; hybrid pushdowns use DataFusion, Spark and Iceberg REST catalogs; an MCP implementation plus an AIDB extension runs Neotron‑45B embeddings and agentic tool calls for text‑to‑SQL and index/observability actions.
AI-written summary · about
chapters (13)
- 0:00Welcome and introduction
- 1:51Transactional vs insight workloads
- 4:00New data modalities for AI
- 6:13Three strategies for Postgres
- 9:02Organic approach with extensions
- 12:17Vector search and GPU acceleration
- 18:37Hybrid strategy with external engines
- 21:23LLMs in Postgres for embeddings
- 23:32Augmenting Postgres with MCP
- 24:57Agentic AI explained
- 31:25Demo of Postgres MCP tools
- 42:30EDB’s Postgres AI platform
- 45:41Q&A session
related talks
34mAnalytics and AI with PostgreSQL - Mission Possible!
37mRound Table Discussion
25mPostgres and the Artificial Intelligence Landscape
34mAnalytics and AI with PostgreSQL - Mission Possible!
45mPostgres and the Artificial Intelligence
01h 00mPostgreSQL and Artificial intelligence
27mPutting AI in Production
44mIntegrating AI with Postgres...