AIPerf

AIPerf is a comprehensive benchmarking tool that measures the performance of generative AI models served by your preferred inference solution.

https://github.com/ai-dynamo/aiperf

EuroPython 2026 Videos

The videos from the EuroPython 2026 talks are now available.

https://www.youtube.com/playlist?list=PLd3Y9yzyC5Uo

JupyterGIS 0.16: New visualization capabilities, collaborative Story Maps, and more

JupyterGIS 0.16 adds richer geospatial visualization with collaborative Story Maps, openEO layers, lazy Xarray rendering, new symbology, and support for GeoZarr and GeoPackage. The release also expands real-time collaboration to vector editing and introduces an R API alongside the existing Python workflow.

https://blog.jupyter.org/jupytergis-0-16-new-visualization-capabilities-collaborative-story-maps-and-more-03e6b78bacc0

Reinforcement Learning for LLMs: The Complete Guide

This is a comprehensive guide to reinforcement learning for LLMs, starting from first principles and building through the policy-gradient algorithms used in modern LLM training. It covers RLHF, verifiable rewards, REINFORCE, PPO, and the newer RL methods being used to improve reasoning, agents, reliability, and efficiency.

https://cameronrwolfe.substack.com/p/llm-rl

DBLift

DBLift is a Python-native database migration toolkit that brings application-style safety to schema changes with validation, dry runs, SQL previews, rollback support, and CI/CD integration.

https://github.com/dblift/dblift

How to Set Up Python for Machine Learning (The Right Way)

AI-generated ML code can run successfully while still containing subtle mistakes that waste hours or days of training. The video shows how to set up a reliable Python ML workflow with the right tools, validation steps, and verification habits to catch problems before expensive training runs.

https://www.youtube.com/watch?v=cAtKtvmds1o

Nifty Django Feature: Counting on Multiple Columns

Django’s Count expression only works on a single column, which makes counting distinct combinations such as unique pet-vet pairs awkward. A custom CountSubquery shows how Django’s expression system can generate the needed SQL while staying inside the ORM instead of dropping to raw SQL.

https://www.better-simple.com/django/2026/08/19/nifty-feature-counting-on-multiple-columns/

ai-job-search

The job search that runs on your machine. AI job application framework built on Claude Code: evaluate postings, tailor CVs, write cover letters, prep interviews. Fork it and own it.

https://github.com/MadsLorentzen/ai-job-search

Speculative Programmatic Tool Calling

Speculative programmatic tool calling (sPTC) speeds up AI agent harnesses by pre-launching tool calls from partially generated code instead of waiting for the model to finish generating. The approach overlaps tool execution with token generation and can also parallelize independent calls, acting like a lightweight JIT compiler for agent-generated programs.

https://alexzhang13.github.io/blog/2026/spec-ptc/

pentestkit

Autonomous multi-agent pentest framework that plans, exploits, verifies with required proof, assigns CVSS scores, and generates client-ready reports. It achieved 104/104 (100%) on the XBOW validation benchmarks and is powered by Kimi K3.

https://github.com/lordx64/pentestkit

Learn Vectorized Thinking in Python Through Examples

In this article, you will learn how to think in terms of vectorized operations using NumPy, replacing slow Python loops with efficient array-level computations.

https://machinelearningmastery.com/learn-vectorized-thinking-in-python-through-examples/

VoiceStudio

VoiceStudio is the open-source, fully local ElevenLabs alternative for voice cloning, voice design, video dubbing, dictation, transcription, and audiobook creation in 646 languages.

https://github.com/debpalash/VoiceStudio

Geolocating Random Islet Image Using Geometry & CUDA GPU Programming

The author solved Sofia Santos’s Gralhix 004 OSINT challenge using a custom geometric pipeline and CUDA GPU acceleration to filter 80.7 million candidate landmass triplets down to 26 locations. The analysis successfully geolocated the image to Oan Resort in Micronesia at 7°21'48.4"N, 151°45'20.7"E, with the camera facing northwest.

https://yassa9.github.io/osint/gralhix-004/

fastaddress

Fastaddress is a Python package that keeps the familiar usaddress API while moving its CRF runtime to Rust for much faster US address parsing. It delivers 11.3x higher single-core throughput, scales to 360K+ addresses per second on eight threads, and matches usaddress output across 20,738 real addresses.

https://github.com/vinvomero/fastaddress

VectorSmith

VectorSmith turns vector database collections into LLM tools and exposes them through an MCP server. Using the existing collection schemas and a simple YAML configuration, you can make your vector DB data available to AI agents without writing custom tool and MCP integrations from scratch.

https://github.com/kjgpta/vectorsmith