#python
Mini-SGLang is a compact, easy-to-read inference framework (~5,000 Python lines) that runs and serves large language models with high speed using optimizations like radix cache, chunked prefill, overlap scheduling, tensor parallelism, and FlashAttention/FlashInfer kernels. It’s CUDA-dependent, quick to install from source, and can launch an OpenAI-compatible API or interactive shell for single- or multi‑GPU serving, letting you test or deploy models (e.g., Qwen, Llama) with low latency and scalable throughput. Benefit: you get a transparent, modifiable engine to deploy fast, efficient LLM inference for development, benchmarking, or production use.
https://github.com/sgl-project/mini-sglang
#python
**ty** is a super-fast Python type checker and language server built in Rust by Astral (makers of uv and Ruff). It's 10-100x faster than mypy or Pyright, with rich error messages, IDE features like auto-complete and hover help, and support for big projects or partial typing. Try it via `uvx ty check`. This helps you catch bugs early, code faster with real-time feedback, and boost productivity in editors like VS Code.
https://github.com/astral-sh/ty
#python
You can access a free, detailed global dataset called the Global Building Atlas, which includes 2D building shapes, heights, and simple 3D models (LoD1) for 2.75 billion buildings worldwide, including areas often missing in other maps like Africa and South America. This data is very accurate, with a fine 3x3 meter resolution, and can be used in GIS software or downloaded fully. It helps with urban planning, disaster risk assessment, climate adaptation, and monitoring sustainable development goals by showing where people live and how cities grow. The dataset and related code are openly available for research and practical use.
https://github.com/zhu-xlab/GlobalBuildingAtlas
#python
cuTile Python is a new programming tool from NVIDIA that lets you write GPU programs in Python more easily and efficiently. It uses a tile-based model, where you work with chunks of data called tiles, making your code portable across different NVIDIA GPUs without needing to rewrite it for each hardware generation. cuTile automatically uses advanced GPU features like tensor cores and memory accelerators, so you get high performance without complex coding. You can install it via pip, and it requires CUDA Toolkit 13.1+ and Python 3.10+. This helps you develop faster, future-proof GPU applications with less effort.
https://github.com/NVIDIA/cutile-python
#python
Foundry is a toolkit that helps design new proteins using powerful AI models. It includes tools to generate protein structures (RFD3), predict how they fold (RF3), and design amino acid sequences that will form those structures (ProteinMPNN/LigandMPNN). All models work together using a common framework for handling molecular structures, making it easier to go from idea to designed protein. The benefit is that it gives a complete, flexible system to create custom proteins for research, medicine, or biotechnology, with clear instructions and examples to get started quickly.
https://github.com/RosettaCommons/foundry
#python
Claude Quickstarts offers ready-made projects that help you quickly build AI applications using the Claude API. You can create tools like a customer support agent, financial data analyst, computer control demo, or an autonomous coding agent by following simple setup steps and using your Claude API key. These projects come with clear instructions and can be customized to fit your needs, saving you time and effort in development. This helps you start building powerful AI apps faster and learn how to use Claude’s advanced features effectively. You also get access to helpful resources, community support, and opportunities to contribute improvements.
https://github.com/anthropics/claude-quickstarts
#python
VibeVoice is an open-source AI tool that creates natural-sounding, expressive audio with up to four different voices, perfect for making podcasts, audiobooks, or long conversations. It keeps each speaker’s voice consistent and handles smooth turn-taking, making the audio sound realistic and engaging. The tool can generate speech in English and Chinese, and even adds spontaneous emotion or singing. It’s free to use and helps creators produce high-quality audio quickly, but should be used responsibly to avoid misuse.
https://github.com/microsoft/VibeVoice
#python
The Social-Engineer Toolkit (SET) is an open-source penetration testing framework created by TrustedSec that helps security professionals test organizational defenses through social engineering attacks. SET provides pre-built attack vectors for phishing, credential harvesting, and website cloning, allowing testers to simulate realistic threats quickly and effectively. The toolkit runs on Linux and Mac OS X, with easy installation via pip or package managers. By using SET with proper authorization, security teams can identify human vulnerabilities in their defenses, understand how employees respond to social engineering tactics, and implement stronger security awareness training to protect against real-world attacks.
https://github.com/trustedsec/social-engineer-toolkit
#python
You can use an AI-powered call center solution built with Azure and OpenAI GPT to automate phone calls for tasks like insurance claims, IT support, and customer service. This system handles calls in multiple languages, streams conversations in real-time, resumes after disconnections, and stores data securely. It uses advanced AI models to understand complex information, manage sensitive data safely, and customize conversations to your needs. The solution scales easily on Azure, offers call recording, human fallback, and brand-specific voices, improving customer experience and reducing costs by automating routine calls while keeping quality and compliance high. This helps you provide 24/7 support efficiently and with personalized service.
https://github.com/microsoft/call-center-ai
#python
This project teaches you how to build a real-world AI research assistant that automatically finds, reads, and answers questions about academic papers using a technique called Retrieval-Augmented Generation (RAG)[1][2][3]. RAG works by first searching for the most relevant information from a large collection of documents, then using a language model to generate clear, accurate answers based on that information—this means you get answers that are up-to-date and grounded in real sources, not just what the AI remembers from its training[1][2][3]. The course is hands-on: each week, you add a new piece, starting with setting up the technical infrastructure, then building automated data pipelines to fetch and process papers, adding powerful search tools (first with keywords, then with AI-powered semantic search), and finally connecting everything to a local AI model that can chat with you and explain complex topics in simple language. By the end, you’ll have a working system you can use to quickly find and understand research papers, and you’ll gain the skills to build similar AI tools for any field—all while learning the best practices used by professional engineers. The main benefit is that you get practical, production-ready AI skills and a tool that makes research faster and more reliable, with answers you can trust because they come directly from the latest papers.
https://github.com/jamwithai/arxiv-paper-curator
#python
You can use Tinker and Tinker Cookbook to easily fine-tune large language models (LLMs) for your specific needs without managing complex training infrastructure. Tinker handles distributed training and uses efficient LoRA adapters to reduce costs and speed up customization. The Cookbook offers ready-made examples and tools for tasks like chat, math reasoning, and reinforcement learning, helping you quickly build and improve AI models. This means you can create AI that better fits your domain, runs faster, and follows your rules, all while saving time and computing resources. It’s great for researchers, developers, and teams wanting powerful, flexible AI customization.
https://github.com/thinking-machines-lab/tinker-cookbook
#python
This project offers free, open-source AI agents designed to help with trading research and automation. It includes tools that can analyze strategies, backtest ideas, monitor markets, and manage risk using advanced AI models. The main benefit is that it lets you test and improve trading strategies safely with historical data before using real money, helping you make smarter decisions and avoid common mistakes. Always remember that trading carries risk and no tool can guarantee profits.
https://github.com/moondevonyt/moon-dev-ai-agents