Scikit-learn 1.9

Scikit-learn 1.9 release is out, and it comes with solid improvements to many existing estimators, making them faster, more stable, handling missing values, adding GPU support… The release also enhances the estimator displays in notebooks, and introduces a callback mechanism that opens the door to progress bars or advanced monitoring of convergence.

https://blog.scikit-learn.org/updates/release-1-9/

Plugins case study: Pluggy

The post explains how Pluggy works as a reusable Python library for building plugin systems, centered on hooks that hosts define and plugins implement. It also highlights the design tradeoffs Pluggy handles well, including registration, signature validation, ordering, and hook wrappers, while leaving plugin discovery largely up to the application.

https://eli.thegreenplace.net/2026/plugins-case-study-pluggy/

PyData London 26

The videos from PyData London 2026 are now available.

https://www.youtube.com/playlist?list=PLGVZCDnMOq0rFQykYJg7t441AEpN4SszE

defending-code-reference-harness

Skills for threat modeling, scanning, triage, patching, plus an autonomous scanning harness you can customize.

https://github.com/anthropics/defending-code-reference-harness

tracesage

tracesage adds local, zero-infra tracing to LangChain/LangGraph agents in two lines, it captures every chain, tool, and LLM call to SQLite and shows the run as a live graph and timeline in your browser. Open source, pip install, MIT licensed.

https://github.com/kjgpta/tracesage

BugHunter

AI-powered bug bounty hunting from your terminal - recon, 20 vuln classes, autonomous hunting, and report generation. All inside Claude Code.

https://github.com/shuvonsec/claude-bug-bounty

Showcasing allauth IdP: build an MCP server

This article demonstrates how to build a Model Context Protocol (MCP) server using Django and django-allauth, with OpenID Connect (OIDC) authentication handled by the allauth.idp package. It shows how MCP hosts such as Claude can dynamically register and authenticate clients without relying on separate third-party identity providers.

https://allauth.org/news/2026/05/idp-demo-mcp-server/

Libraries for your Python Polars workflows

The post highlights a set of Python libraries that work natively with Polars, letting you stay in the Polars ecosystem for validation, tables, visualization, and LLM-assisted analysis.

https://opensource.posit.co/blog/2026-06-04_libraries-for-python-polars/

How I made dependency injection in Python 130× faster: from reflection to compiling the graph

The post shows how a Python dependency injection container was optimized from ~53 µs to 0.40 µs per resolve by caching plans, eliminating unnecessary checks, and compiling dependency graphs into generated code. Along the way, it demonstrates practical performance engineering techniques, including code generation, common-subexpression elimination, fuzz testing, and reproducible benchmarking

https://vshulcz.hashnode.dev/python-dependency-injection-130x-faster

How to build your first Scrapy extension

This tutorial uses a simple audio notification plugin to explain how Scrapy extensions work, covering signals, lifecycle hooks, settings, testing, and packaging. It provides a practical introduction to extending Scrapy with custom functionality while showcasing common patterns used across the ecosystem.

https://www.zyte.com/blog/how-to-build-your-first-scrapy-extension/

Array API adoption: what to do with compiled code

The post examines how Array API compatibility and JIT compilation can modernize legacy scientific Python code for GPUs and multicore CPUs. The results suggest developers can achieve significant performance gains across hardware platforms without maintaining specialized accelerator-specific code.

https://labs.quansight.org/blog/array-api-aot-jit

The Fastest Python Struct?

The article explores the performance tradeoffs of Python's various struct-like types, focusing on startup and type-definition costs rather than runtime business logic. Extensive benchmarks reveal that NamedTuple sits in the middle of the pack, while msgspec and C-backed implementations deliver the fastest type creation and startup times.

https://www.crumpledpaper.tech/2026-06-21-python-struct-profiling/

Write a coding agent from first principles

This tutorial will show you how to create your own coding agent from first principles. By doing so, you'll understand how coding agents work under the hood.

https://mathspp.com/blog/write-a-coding-agent-from-first-principles

PixelRAG

PixelRAG replaces traditional text-based web RAG with a pixel-native approach that retrieves and reasons over webpage screenshots. By preserving visual structure and using vision-language models, it aims to improve retrieval accuracy and web understanding.

https://github.com/StarTrail-org/PixelRAG

Wagtail as Django admin on steroids

Wagtail can be used as a drop-in replacement for Django Admin, providing a more polished UI and powerful customization capabilities. The article demonstrates how developers can adopt Wagtail incrementally while retaining their existing Django architecture.

https://timonweb.com/wagtail/wagtail-as-django-admin-on-steroids/

NVlabs / Eagle

Eagle: Frontier Vision-Language Models with Data-Centric Strategies

https://github.com/NVlabs/Eagle

Optocam Zero

Optocam Zero is a Raspberry Pi Zero based compact digital camera made using off the shelf components.

https://github.com/dorukkumkumoglu/optocamzero

WinPodX

Windows pod system for Linux.

https://github.com/kernalix7/winpodx

NVIDIA / skills

AI agent skills published by NVIDIA

https://github.com/NVIDIA/skills