From Queries to Agents: The Next Era of Data Retrieval on PostgreSQL
summary
Argues Postgres becomes an agent backend by combining MCP servers, schema/semantic/relationship/safety context correction, and in‑DB blended retrieval using pg_fts (BM25), pgvector+DiskANN, Apache AGE graph queries and the azure_ai extension. Demoed Product Copilot where vector‑only search returns four items over $100, while the blended retrieval pipeline enforces the <$100 filter, re‑ranks results and returns relevant bookshelves. Presents HorizonDB as a single Postgres offering with MCP connectors and a roadmap toward a unified ai.search() API.
AI-written summary · about
chapters (12)
- 0:00Music & Introduction
- 0:22Speaker introduction: sequel to last year's RAG talk
- 1:02When AI agents go wrong: real horror stories
- 3:08Where RAG falls short for complex queries
- 4:29Agent architecture: multi-step, multi-tool retrieval
- 5:28Model Context Protocol: usb-c for your AI
- 6:46Failure modes: hallucination, semantics, guardrails
- 8:44Context correction: four layers for trustworthy agents
- 12:30Azure HorizonDB and four retrieval modes in one db
- 16:54Demo: Blended retrieval in Product Copilot
- 20:39What's next: unified AI.search() in HorizonDB
- 22:21Resources, GitHub repo, and thanks
related talks
53mModern PostgreSQL for AI...
34mAnalytics and AI with PostgreSQL - Mission Possible!
48mThe power of AI and Copilot for Azure Databases
12mThe mechanics of integrating AI and vector search into Azure Database for PostgreSQL
34mAnalytics and AI with PostgreSQL - Mission Possible!
50mFrom Transactions to Intelligence: Postgres for Open AI & Analyti
44mPostgres and OpenAI - a match made in Azure
37mRound Table Discussion