Building AI Applications the Pydantic Way
summary
Shows a concrete toolkit: Pydantic AI agent plus Pydantic Logfire and an MCP server called MCP run Python. Demonstrations extract JSON-typed outputs, trigger a validation failure that causes an agent retry (two model calls), compare models side-by-side, and add long‑term (Postgres) vs short‑term (message context) memory; an eval suite improves pass rate to ~83% after a prompt change and traces costs (example run ~£0.15). If you need type‑safe agent patterns, observability and sandboxed Python execution, this is pragmatic and hands‑on.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 1:26Welcome and Pyantic overview
- 3:36AI applications challenges
- 6:03Agents and agent loops
- 7:52Structured data extraction
- 9:39Agents and validation
- 12:48Tool registration and type safety
- 15:35Long-term memory
- 18:45Short-term memory and context
- 22:16Model Context Protocol
- 26:26Sandboxed Python execution
- 31:06Browser control with MCP
- 36:03Open standards
- 37:49Pydantic Evals introduction
- 39:34Building eval datasets
- 43:25Evaluators and validation
- 45:14Running evals and results
- 49:45Comparing models
- 52:17Pydantic AI roadmap
- 53:39Q&A
related talks
30mGenerative AI Monitoring with PydanticAI and Logfire
01h 21mAgent Optimization with Pydantic AI: GEPA, Evals, Feedback Loops
27mHow to Build AI Agents and Keep Your Sanity
30mBuilding Reliable AI Agents: An Open Source Approach To Evaluation and Observability
12mHuman seeded Evals
29mDemystifying AI Agents with Python Code
52mCut the Hype and Build Your Own AI Agents
48mDyalog and AI