Getting Started with AI Agents in Python
Workshop Details
Practical Information
Laptop, internet connection, git
You need to know how to run commands and edit code. You don't need to know Python but you must know how to code in general.
About this Workshop
Whether you know Python or not, this workshop will help you set up a ready-to-use environment in minutes and quickly get hands-on.
The workshop is structured around two concrete mini-projects designed to explore different ways of building and using agents.
1) Data Analysis and Visualization
We'll start with a deliberately naive approach using pydanticai, providing data directly to the agent. We'll then move to smolagents, with a CodeAgent capable of writing and executing code to produce more advanced analyses and visualizations. Finally, we'll return to pydanticai, this time enriched with tools, allowing the agent to perform more robust data analyses and compare the different approaches concretely.
2) Creating Tools (MCPs)
In this second mini-project, we'll build tools (MCPs) allowing an agent to search for products on a well-known Swiss e-commerce website. The goal will be to answer complex queries such as: "Show me 5 CPUs compatible with my motherboard, under 200 CHF and in stock."
Finally, we'll compare our solution with more generic tools like browser-use or ChatGPT (in agent mode and classic mode).
Speakers & Organizers
William Droz
Senior Research and Development Engineer
Idiap Research Institute
William Droz is a Senior Research and Development Engineer at the Idiap Research Institute, where he designs and deploys robust, maintainable AI systems bridging research and real-world applications. He specializes in Python (scikit-learn, Polars, PyTorch, FastAPI), works primarily on GNU/Linux, and has a strong interest in Rust, focusing on projects involving agents, LLMs, data analysis, and production-ready tools.