9. SaarPython Meetup

Dear All,

We are getting ready for the 9. SaarPython Meetup (see the agenda below).

Send us your talk proposals as soon as possible. Preferably, lightning talks with a length of 10-15 minutes (incl. short Q&A).

Thanks again to natif.ai GmbH for hosting the event.

Refreshments and snacks will be provided during the event. The capacity of the venue is limited, so make sure to RSVP early enough and to un-RSVP if you can't make it to the event.

Let's get together for another round of interesting Python talks!

Yves

***

Alastair Burt - Declarative, Reproducible Python with Nix
This talk will explain what the Nix package management system is, why it is declarative and why it leads to reproducible systems. It will show various ways in which Nix can be used for general package management and how these carry over to reproducible Python applications.

Mohamad Hammoud - Would You Approve This MR? Python Edition
In this interactive session, we’ll review short Python snippets as if they were merge requests generated by an AI assistant. Your job: would you approve this MR? Safe, unsafe, or… it depends?
We’ll explore subtle security boundaries in everyday Python code, subprocess calls, file handling, URL validation, deserialization, caching, and more inspired by real-world vulnerabilities. For each unsafe example, we’ll also look at how it should be implemented properly.

Sebastian Kalkowski - Hands-free Coding from the Sofa
The Model Context Protocol (MCP) lets you extend AI coding assistants with custom capabilities — and it turns out that giving them a voice is surprisingly straightforward. In this talk I’ll show how to build a fully local voice interface for Claude Code, turning it into a hands-free pair programming partner. We’ll go from zero to live demo in under 100 lines of Python. No keyboard required.

Felix Krüger - A Lightning Introduction: Taming the PyTorch Training Loop with PyTorch Lightning
Training models in plain PyTorch offers maximum flexibility but requires manually implementing training loops, logging, checkpointing, and scaling infrastructure. PyTorch Lightning introduces a lightweight wrapper around PyTorch that cleanly separates model definition from training infrastructure, reducing boilerplate while preserving flexibility. In this talk, we will present a practical side-by-side comparison of raw PyTorch and PyTorch Lightning, and demonstrate how these abstractions help structure real-world ML systems.