Python

dchristopoulos/jev-aita

Independent benchmark tests TypeSafe's Jev against Sonnet 5 on AITA verdicts — An independent, self-funded test of TypeSafe's new 'System One' model Jev against Sonnet 5, GPT-5 nano and local LLMs, predicting Reddit AITA verdicts on 770 posts. Reports Brier scores, latency and cost, and checks TypeSafe's speed claims—useful for anyone evaluating fast probability-only models versus full LLMs.

Benchmark of TypeSafe's Jev against Sonnet 5, GPT-5 nano and local LLMs on 770 Reddit AITA verdicts: Brier scores, latency and cost

shirok-tech/lakehouse-sound-festival-2026-paf-demo

A fully synthetic lakehouse demo built for Oracle's Autonomous AI stack — This repo offers a reproducible hands-on dataset for Oracle Autonomous AI Lakehouse, Select AI and Private Agent Factory, modeling a fictional music festival with 137,535 synthetic records, Parquet files and PDFs. It now adds an Ontology sequel with SQL property graphs and knowledge-graph views, useful for teams testing KG-guided RAG and agent workflows on Oracle's platform.

Reproducible Oracle Autonomous AI Lakehouse and Private Agent Factory demo dataset for Lakehouse Sound Festival 2026

OpenBMB/MiniCPM-V

MiniCPM-V 4.7 brings MoE multimodal AI to phones and edge devices — MiniCPM-V 4.7 adds a 35B-A3B MoE variant plus a 1B model, claiming better accuracy-per-latency than Qwen3.6 35B-A3B and Qwen3.5 0.8B. It's the first open-source MoE model said to run smoothly on-device, aimed at developers wanting local image, video and document understanding without cloud inference.

See, hear, and interact on your local devices.

anthropics/financial-services

Claude agents built for investment banking and wealth management desks — A collection of reference Claude agents, skills, and connectors for IB, equity research, private equity, and wealth management workflows. Deploy as Cowork plugins or via the Managed Agents API from one source; outputs like memos and models are staged for human review, not automated decisions.

GantMan/nsfw_model

A trained Keras classifier for flagging NSFW images — A Keras/Inception V3 model trained on 60+ GB of images to classify content into drawings, hentai, neutral, porn, and sexy categories with 93% accuracy. Offers a Python API and CLI for single or batch predictions, and powers the NSFW JS project for client-side content moderation.

Keras model of NSFW detector

pytorch/pytorch

The Python tensor library behind modern deep learning research — PyTorch combines NumPy-like GPU-accelerated tensors with a tape-based autograd system for building dynamic neural networks. Its imperative, Python-first design lets researchers debug and extend models with familiar tools like NumPy, SciPy, and Cython, making it a core choice for ML experimentation.

Tensors and Dynamic neural networks in Python with strong GPU acceleration

JEOresearch/EyeTracker

A Python-based 3D pupil tracker refined from an existing algorithm — This project reworks the pupil-detection algorithm from Yuta Itoh's 3D-Eye-Tracker, simplifying and optimizing it in pure Python. It returns a fitted ellipse for the pupil and includes both a robust mode and a faster lite mode for well-lit setups, useful for anyone building DIY eye-tracking hardware.

A lightweight and robust Python eye tracker

getsentry/sentry

Sentry adds AI-assisted debugging alongside tracing and logs — Sentry's self-hosted platform now bundles issue tracking, tracing, session replay, uptime monitoring, and a Seer AI debugging assistant in one dashboard. It targets teams wanting a self-managed, fair-source alternative to hosted error-tracking SaaS, with SDKs covering most major languages and frameworks.

Developer-first error tracking and performance monitoring

element-hq/synapse

Synapse, the Matrix homeserver, now shipped via Element Server Suite — Synapse is Element's open source Matrix homeserver, written in Python/Twisted with Rust components, licensed AGPL or commercially. It's now distributed through Element Server Suite (ESS), with free Community, Pro, and a Germany-specific TI-Messenger edition for healthcare compliance, making deployment and support options clearer for operators.

Synapse: Matrix homeserver written in Python/Twisted + Rust

abus-aikorea/voice-pro

One Gradio UI for TTS, Voice Cloning, Whisper, and Dubbing — Voice-Pro bundles Edge-TTS/Kokoro TTS, zero-shot voice cloning (E2/F5-TTS, CosyVoice), Whisper-based transcription, YouTube download, Demucs vocal isolation, and multilingual translation into one Gradio web app. It targets creators and developers who want an end-to-end pipeline for subtitling, dubbing, and podcast/audiobook production without stitching separate tools together.

Gradio WebUI for creators and developers, featuring key TTS (Edge-TTS, kokoro) and zero-shot Voice Cloning (E2 & F5-TTS, CosyVoice), with Whisper audio processing, YouTube download, Demucs vocal isolation, and multilingual translation.

Diolinux/PhotoGIMP

A Photoshop-style skin and workflow patch for GIMP 3 — PhotoGIMP reorganizes GIMP's tool layout, shortcuts, and canvas settings to match Adobe Photoshop, easing the switch for former Photoshop users. It now targets GIMP 3+, includes a custom splash screen, and has README translations in ten languages, showing active community maintenance.

A Patch for GIMP 3+ for Photoshop Users

Amal-David/awesome-jev

A curated map of Jev demos, SDKs, and agent tooling — This awesome-list gathers demos, SDKs, browser/desktop tools, and reading material around Jev, TypeSafe's probability-based yes/no judgment model. It's aimed at developers building typed classifiers or agents, with source links and notes on setup, privacy, and evidence limits rather than just hype.

Jev demos, projects, SDKs and skills, with source links and a curated X gallery.

Significant-Gravitas/AutoGPT

AutoGPT now ships a visual builder for self-hosted AI agents — AutoGPT lets users describe a desired outcome or design every step visually, then run agents on demand, on a schedule, or via triggers. With over 185,000 stars and a self-host option, it suits developers and non-coders who want autonomous task completion without relying on a hosted platform alone.

AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.

sqlfluff/sqlfluff

A dialect-aware SQL linter built for ELT and dbt workflows — SQLFluff lints and auto-fixes SQL across many dialects (BigQuery, Snowflake, MySQL, etc.) and handles Jinja/dbt templated code. It targets ELT pipelines and dbt users who need consistent SQL style without manual cleanup, with a VS Code extension for editor integration.

A modular SQL linter and auto-formatter with support for multiple dialects and templated code.

django/django

A mature Python framework built for pragmatic, fast app development — Django bundles ORM, templates, views and admin tooling into one cohesive Python framework, letting developers build and deploy web apps without stitching together separate libraries. Extensive official docs, tutorials and an active Discord/Forum community make it approachable for beginners and teams on deadlines alike.

The Web framework for perfectionists with deadlines.

allenai/olmocr

A 7B VLM toolkit that turns PDFs into clean Markdown for training data — olmOCR converts PDFs, PNGs and JPEGs into readable Markdown, preserving reading order, tables, equations and handwriting while stripping headers and footers. Backed by a 7B VLM (GPU required), the new v0.4.0 release boosts its olmOCR-bench score via synthetic data and RL training, at under $200 per million pages.

Toolkit for linearizing PDFs for LLM datasets/training

feder-cr/invisible_playwright_mcp

A Playwright MCP agent built to slip past anti-bot defenses — This MCP server drives a stealth Firefox browser that AI agents like Claude Code or Gemini CLI can use to browse without triggering anti-bot systems or captchas. It plugs into assistants via MCP for tasks like job hunting, scraping, and general computer use.

Playwright MCP server undetected by anti-bots and captchas: AI agent browses the web on anti-detect stealth Firefox, Python, undetected browser automation, scraping, computer use.

dfd-tud/deda

Decode and erase the hidden yellow tracking dots laser printers leave — DEDA extracts, decodes and anonymises the yellow tracking dots that most color laser printers embed in printouts, exposing serial numbers and other identifying data. It offers a GUI and CLI tools for parsing scans, comparing prints to spot a specific printer, and analysing unrecognised dot patterns—useful for forensic researchers, journalists and privacy-conscious users.

net4people/bbs

A GitHub-issues forum where circumvention researchers compare notes — Net4People BBS uses GitHub issues as a multilingual discussion board for developers and researchers working on Internet censorship circumvention. Anyone can open an issue in their own language to share techniques, tools, or findings, with translation help from the community.

Forum for discussing Internet censorship circumvention

snowflakedb/snowflake-connector-python

Python DB API driver for Snowflake reaches 5.0.0, then moves on — Version 5.0.0 marks the last release here: future development shifts to snowflakedb/drivers. The connector implements PEP 249 DB API 2.0 with no JDBC/ODBC dependency, letting Python apps query Snowflake directly on Linux, Mac, and Windows with Python 3.10+.

Snowflake Connector for Python

python/cpython

CPython's main branch now building toward 3.16.0 alpha 1 — This is the official reference implementation of Python, currently tracking version 3.16.0 alpha 1 development on the main branch. Useful for contributors following the Developer's Guide, testers of upcoming language features, and anyone building Python from source rather than using a packaged release.

The Python programming language

chardet/chardet

Python character encoding detector

fhshaik/typesafe-mario

Letting a language model play Mario using only structured JSON state — Instead of feeding screenshots to an AI, this project converts NES RAM and telemetry into compact object-centric JSON (Mario's motion, enemies, terrain) for TypeSafe's Jev model to choose controller inputs. A live dashboard shows action probabilities, confidence, and latency, making it useful for anyone exploring structured-state game agents.

A TypeSafe/Jev agent that plays Super Mario Bros. from structured emulator state.

General-Instinct/InstinctFlash

Fused FP8 kernels push robotics model inference up to 33x faster — InstinctFlash is a serving runtime for robotics models like LingBot-VA, pi05 and GR00T N1.7, with purpose-built FP8 engines on Jetson Thor plus RTX 4090/5090 support. Benchmarks claim up to 33.78× speedup with no observed task-performance loss, aimed at robotics engineers deploying real-time inference.

High-Performance Serving Runtime for Robotics Models

dukesun99/Corpus2Skill

Turning document corpora into navigable skill trees for agents — Official EMNLP 2026 Findings code for Corpus2Skill, which compiles enterprise document corpora into a hierarchy of Anthropic Skills. Agents navigate SKILL.md/INDEX.md files and fetch documents on demand at query time, replacing vector-search or BM25 serving. Useful for researchers and builders exploring alternatives to embedding-based RAG for enterprise QA.

Official Findings of EMNLP 2026 implementation of Corpus2Skill: compile a document corpus into a navigable skill hierarchy that LLM agents explore at query time, with document lookup instead of a serving-time vector-search service.

grafana/o11y-bench

Benchmarking LLM agents against a real Grafana observability stack — o11y-bench evaluates LLM agents on observability and SRE tasks, running them against live Prometheus, Loki, and Tempo via the Harbor framework. It grades results with deterministic checks plus rubric criteria, producing HTML reports and a leaderboard, useful for teams testing agentic AI on real-world monitoring workflows.

Open benchmark on observability tasks built on Harbor

brumar/chess-postmortem-skills

Turn a chess game plus your spoken thoughts into a narrated post-mortem — These Claude Code skills combine a Stockfish-backed annotated PGN, HTML viewer, and narrated video to explain your chess mistakes in plain language. It even matches a recorded voice memo of your in-game thinking to each move via whisper.cpp transcription and clock times, then answers your own questions with the engine.

Claude Code skills for chess post-mortems: Stockfish-backed annotated PGN, HTML viewer, narrated video

lowenbjer/claude-terse

A Claude Code plugin that strips the essay tone out of replies — This plugin targets Claude Code's tendency to answer every question with essays, metaphors, and

For those who struggle with how Claude speaks to them. No more essays. No more slogans, metaphors, "it's not X, it's Y". This plugin makes Claude speak normally.

bertaye/agentic-cuda-optimizer

An LLM agent that iteratively rewrites and benchmarks CUDA kernels — A LangGraph-driven agent that generates, compiles, and benchmarks CUDA kernels in a loop, using a C++/NVRTC harness for correctness checks and Nsight Compute counters for optimization clues. It tunes both kernel code and launch configs, keeping the fastest validated version. Useful for GPU developers exploring automated kernel tuning.

A langgraph based workflow with a C++ CUDA harness to optimize CUDA kernels

adrianoz1/ai-video-cut

Turning YouTube videos into viral vertical clips with AI — This Python tool automates the whole short-form clip pipeline: downloading YouTube videos, transcribing with Whisper, finding highlight moments via GPT-4o-mini, and rendering TikTok-style word-by-word captions in vertical 9:16 format. Useful for creators who want to repurpose long videos into Shorts/Reels without manual editing.

Processador Automático de Vídeos do YouTube