Simplify End-To-End MLOps with PostgresML
summary
Claims that running ML inside Postgres with PostgresML is far faster and simpler than Python microservices, citing 10x faster end-to-end for embedding+RAG vs remote OpenAI calls. Implements train/deploy/predict SQL functions, in-DB Hugging Face Transformers (Rust extension, Python bridge), model RAM/GPU caching, and a pgcat proxy for sharding. Demonstrates a nearest-neighbor search over 5 million reviews in 70ms and supports embed+LLM in-DB workflows; code is MIT open source.
AI-written summary · about
chapters (8)
- 0:00The case for in-database ML
- 5:00Data modeling and scale
- 17:09Scaling with Postgres ML
- 20:36Architectural simplification
- 24:43Streamlining RAG pipelines
- 26:06Core database functions
- 27:26In-database LLM interaction
- 29:08Vector search in practice
related talks
25mPostgres and the Artificial Intelligence Landscape
29mBuilding the Bridge Between DevOps and MLOps in a GenAI world
01h 00mPostgreSQL and Artificial intelligence
10mML-based Log Analysis for Faster Debugging
35mBreakout Session
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
50mFrom Transactions to Intelligence: Postgres for Open AI & Analyti
33mBuilding Resilient (ML) Pipelines for MLOps