How Can You Structure Your Python Script, Optimizing Test Execution and more
with some more interesting news, articles, packages and projects
News
The Python Language Summit 2025
The Python Language Summit 2025 occurred on May 14th in Pittsburgh, Pennsylvania. Core developers and special guests from around the world gathered in one room for an entire day of presentations and discussions about the future of the Python programming language.
DjangoCon Europe 2025 Videos
PEP 685: Comparison of Extra Names for Optional Distribution Dependencies (Final)
PEP 751: A File Format to Record Python Dependencies for Installation Reproducibility (Final)
Articles
Optimizing Test Execution: Running live_server Tests Last with Pytest
It’s a good idea to run slow tests last; this way, you catch failed tests early and avoid putting unnecessary stress continuously on resources like databases and external services. In this article, Tim Kamanin shared how to achieve this by creating a custom marker and a pytest plugin tailored for Django application tests.
# Tim Kamanin
Navigating Imbalanced Datasets with Pandas and Scikit-learn
Imbalanced datasets are a common challenge in real-world machine learning. If not handled properly, models tend to favour the majority class, leading to biased results. In this article, Iván Palomares Carrascosa shared three effective strategies(Inverse Frequency, Undersampling, and Oversampling) to balance such data using Pandas and Scikit-learn.
# Iván Palomares Carrascosa
How Can You Structure Your Python Script?
Many people write Python scripts to automate tasks, but without proper structure, these scripts can quickly become hard to maintain. If you're planning to write a script, check out this Real Python article by Rohit Goswami. Your future self (and your teammates) will thank you. He explained how to organise your code using functions, constants, enums and data classes; improve usability with command-line arguments; logging; and even make scripts self-contained using PEP 723 to handle dependencies inline.
# Rohit Goswami
Beyond Model Stacking: The Architecture Principles That Make Multimodal AI Systems Work
I hope you remember in the Iron Man movie when Tony Stark flew around and JARVIS could understand what was in front of him? That kind of real-time visual understanding feels like science fiction, but Eric Chung wanted to bring it closer to reality. So he built VisionScout, a system combining YOLOv8 for object detection, CLIP for semantic reasoning, Places365 for scene classification and Llama 3.2 for language generation. He shared the full journey, along with insights and implementation details, in his article on Towards Data Science.
# Eric Chung
In this video, Vincent D. Warmerdam and Johnny demoed some of the powerful features of Marimo Notebook and walked through the amazing things you can build with it. It’s a great showcase of how Marimo stands out from traditional notebooks.
# Vincent D. Warmerdam & Johnny
Interesting Packages and Projects to Explore
Dishka - Cute DI framework with scopes and agreeable API
thepi.pe - Get clean data from tricky documents, powered by VLMs
Manim - Animation engine for explanatory math videos
Extractous - Fast and efficient unstructured data extraction. Extractous Python Binding
Bagels - Powerful terminal expense tracker.
Recent Noteworthy Package Releases
python-calamine 0.4.0 - Python binding for Rust's library for reading excel and odf file - calamine
SeleniumBase 4.40.0 - A complete web automation framework for end-to-end testing
pylance 0.31.0 - Python wrapper for Lance columnar format
PyAV 15.0.0 - Pythonic bindings for FFmpeg's libraries
PEFT 0.16.0 - Parameter-Efficient Fine-Tuning (PEFT)
CrewAI 0.140.0 - Cutting-edge framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks
statsig-python-core 0.6.0 - Statsig Python bindings for the Statsig Core SDK
haystack-experimental 0.11.0 - Experimental components and features for the Haystack LLM framework
wandb 0.21.0 - A CLI and library for interacting with the Weights & Biases API.
fastmcp 2.10.0 - The fast, Pythonic way to build MCP servers.
feast 0.50.0 - The Open Source Feature Store for AI/ML
sentence-transformers 5.0.0 - Embeddings, Retrieval, and Reranking
PaddlePaddle 3.1.0 - Parallel Distributed Deep Learning
pillow-heif 1.0.0 - Python interface for libheif library
bleak 1.0.0 - Bluetooth Low Energy platform Agnostic Klient
browser-use 0.4 - Make websites accessible for AI agents
PostHog 6.0.0 - Integrate PostHog into any python application