[Feature Request] Add Pydantic AI to Unit 2: Frameworks for AI Agents #659
Description
Issue Body:
Description
I would like to propose adding Pydantic AI as a featured framework in Unit 2 (Frameworks for AI Agents) or as a dedicated bonus unit.
Pydantic AI is a modern, lightweight, and strictly type-safe framework for building agentic workflows. Given the course's focus on teaching best practices for building reliable agents, Pydantic AI’s approach to "Model-Driven" development is a perfect fit.
Why Pydantic AI?
While the course currently covers excellent frameworks like smolagents and LangGraph, Pydantic AI offers a unique value proposition that aligns with modern Python development:
-
Type Safety & Validation: Leverages Pydantic (the industry standard) to ensure structured outputs are validated at the core.
-
"FastAPI of Agents": It feels familiar to anyone who has built web APIs in Python, reducing the learning curve.
-
Dependency Injection: It has a built-in, type-safe DI system that is perfect for production-grade agentic systems (e.g., passing DB connections or user context).
-
Lightweight & Minimalist: Like smolagents, it avoids over-abstraction, making it easier for students to see "under the hood."
Proposed Content Outline
I suggest a sub-unit (e.g., Unit 2.4) or a Bonus Unit covering:
Introduction to Pydantic AI: The philosophy of type-safe agents.
Defining Structured Outputs: Using Pydantic models to guarantee agent responses.
Tools & Dependency Injection: How to provide agents with external context and capabilities safely.
Streaming & Observability: Real-time responses and integration with tools like Logfire.
Hands-on Lab: Building a simple agent that takes a user query and returns a strictly validated JSON object (e.g., a "Weather Report" or "Financial Summary" agent).
Alignment with Hugging Face
Pydantic AI is model-agnostic and works seamlessly with Hugging Face models via OpenAI-compatible endpoints or local integration. It would complement smolagents by showing students how to scale from "code-first" agents to "schema-first" production systems.
Contribution
I am happy to help draft the content or provide examples for the notebook if the maintainers are open to this addition!