As AI systems evolve from simple chat interfaces into fully autonomous assistants, developers are learning that the most important component is no longer the model. It is the database beneath it. Postgres is emerging as the ideal foundation for trustworthy AI because it already provides the qualities AI systems need: durable memory, expressive retrieval, transactional guarantees, auditing, and safe interfaces for executing real work.
This talk shows how to turn Postgres into a full AI application server by combining Retrieval Augmented Generation (RAG) and the Model Context Protocol (MCP) directly inside the database. We will build a complete, end to end example application, an AI Portfolio Manager, to demonstrate how Postgres can power both the “thinking” and the “doing” of an AI agent.
Attendees will learn how to store and chunk financial documents, generate embeddings with pgvector, combine semantic search with lexical ranking, cache answers, track provenance, and build hybrid retrieval pipelines using nothing more than SQL and Postgres extensions. On the action side, we will expose safe, purpose built functions through MCP so that LLMs can run portfolio queries, generate risk snapshots, simulate rebalances, or submit trades, all within strict roles, policies, and audit logs that prevent misuse.
By the end of the session, you will see how Postgres can unify the entire AI stack: text data, embeddings, metadata, retrieval, safety controls, tool execution, and monitoring. You will walk away with architectural patterns, schema templates, and code samples showing how to embed AI agents directly into your Postgres environment, enabling richer, safer, and more powerful AI applications without introducing yet another database to the stack.
This talk blends conceptual clarity with practical implementation and includes a live demonstration of Postgres acting as the brain, memory, and control surface for an intelligent portfolio management agent.
