Projects

### Overview: 《Deep Agents 实战》是一门基于 LangChain / LangGraph 的中文开源课程，旨在从零构建生产级 AI Agent。课程包含17章图文与实验模板，从Agent的创建开始，逐步介绍任务规划、文件系统、子Agent协作与上下文管理。 ### Key features: - 从零构建生产级AI Agent - 17章图文与实验模板 - 从Agent创建开始，逐步掌握任务规划、文件系统、子Agent协作与上下文管理 - 课程配套17分钟上手指南 - 从0.5起步，推荐使用最新版本0.7.x ### Tech stack: Jupyter Notebook ### Why it matters: 通过本课程，开发者可以快速上手并构建自己的AI代理，掌握生产级AI Agent的核心技术和实战经验。

### Overview: VoxelMorph is a powerful library for learning-based image registration and modeling deformations. It provides a comprehensive toolkit for various applications, including registration of medical images, registration of CT and MRI scans, and more. ### Key Features: - **Image Registration**: VoxelMorph excels at aligning images using deep learning techniques, making it ideal for medical imaging tasks. - **Deformation Modeling**: The library supports modeling deformations using various types of deformations, including affine, elastic, and more. - **Data Loading**: It includes a flexible data loading mechanism that supports multiple formats such as NIfTI, MGZ, and npz files. - **Model Training**: VoxelMorph comes with pre-trained models for various applications, and users can train their own models by customizing the data loading code. ### Tech Stack: Python ### Why it Matters: VoxelMorph is highly effective for tasks that require precise image alignment and deformation modeling, making it a valuable tool in fields such as medical imaging and computer vision. Its versatility and robustness make it a preferred choice for researchers and practitioners alike.

Overview: NOF0 是一个开源的 AI 交易竞技场项目，旨在通过真实加密货币市场数据来验证不同 AI 模型的交易策略。它完整复刻了 NOF1 Alpha Arena，提供了一个以 Prompt 为中心的交易平台，让任何人都可以部署自己的 AI 交易竞技场。 Key features: - 实时展示每个模型的盈亏表现 - 完整开源复刻 NOF1 Alpha Arena 的功能 - 一键部署基础设施，专注 Prompt 策略本身 - 用真实数据和清晰可视化回答 "哪个模型更会赚" 的朴素问题 Tech stack: Go Why it matters: NOF0 为 AI 和加密货币交易领域提供了一个创新的平台，通过实盘竞技和清晰的数据展示，帮助用户验证和比较不同 AI 模型的交易策略，从而推动这一领域的技术发展和市场教育。

Overview: Pluely is an AI assistant that seamlessly integrates into meetings, interviews, and conversations without anyone knowing. It provides real-time answers and transcripts, making it undetectable in video calls, screen shares, and recordings. Key features: - **Listen Mode**: Real-time transcription and answer generation. - **Ask Mode**: Screenshot and code-based responses. - **Dashboard**: Full chat, file, prompt, and settings management. Tech stack: TypeScript, built with Tauri for native performance, and just 10MB in size. Why it matters: Pluely offers a privacy-first solution that works invisibly in video calls, screen shares, and recordings, making it ideal for professionals who want to maintain their professional image.

Overview: zi2zi is an application and extension of the pix2pix model designed to learn Chinese calligraphy through conditional adversarial networks. It uses conditional GANs to generate characters from their corresponding Chinese characters, improving generalization and performance on unseen data. Key features: - Conditional GANs for character generation - Label shuffling to mitigate flatline of discriminator loss - Comparison with ground truth, brush writing, cursive script, and Mingchao style - Interpolation and animation effects Tech stack: Python, TensorFlow, Pillow, numpy, scipy, imageio Why it matters: By leveraging conditional GANs and label shuffling, zi2zi enables more accurate and detailed character generation, making it a valuable tool for learning Chinese calligraphy and improving the performance of conditional GANs on unseen data.

### Overview: Carbon is an open-source manufacturing ERP, MES, and QMS system that integrates MRP, inventory, shop floor, quality, and lot/serial traceability functionalities on a single Postgres schema. It offers a REST API and MCP server, enabling self-hosting or using Carbon Cloud. ### Key features: - **ERP and MES Integration**: Seamlessly combines ERP, MRP, MES, and QMS functionalities. - **Inventory Management**: Tracks inventory levels and manages materials. - **Shop Floor Management**: Controls production processes and shop floor operations. - **Quality Management**: Monitors and tracks quality metrics. - **Lot/Serial Traceability**: Tracks and traces every unit to its source, facilitating lot and serial traceability. ### Tech stack: TypeScript ### Why it matters: Carbon is designed to help hard tech unicorns build and manage manufacturing operations efficiently, combining ERP, MRP, MES, and QMS functionalities in a single system, making it easier to manage materials, run the shop floor, manage quality, and track costs. The system's modular architecture and self-hosting capabilities make it a versatile solution for various manufacturing environments.

Overview: Rhasspy is an open-source, fully offline voice assistant service designed for many human languages, optimized for working with Hermes protocol compatible services, Home Assistant, Node-RED, Jeedom, and OpenHAB. It supports pre-specified voice commands, open-ended speech recognition, and handling of uncommon words and pronunciations. Key features: - Offline operation - Support for many human languages - Optimized for MQTT, HTTP, and Websockets - Pre-specified voice commands - Open-ended speech recognition - Handling of uncommon words and pronunciations Tech stack: Shell Why it matters: Rhasspy is optimized for working with external services via MQTT, HTTP, and Websockets, making it versatile and efficient for integration with various home automation and IoT platforms. Its pre-specified and open-ended capabilities make it suitable for a wide range of use cases, ensuring flexibility and ease of implementation.

Overview: TokenSpeed is a speed-of-light LLM inference engine designed for agentic workloads, offering TensorRT-LLM-level performance and vLLM-level usability. Key features: - **Scheduler**: C++ control plane and Python execution plane - **Kernels**: Pluggable, layered kernel system with a portable public API and centralized registry - **Entrypoint**: SMG-integrated AsyncLLM for low-overhead CPU-side request handling Tech stack: Python Why it matters: TokenSpeed is the first open-source LLM inference engine to separate the control plane from the execution plane, combining strong correctness guarantees with high development velocity. It is backed by the non-profit LightSeek Foundation, offering serious hardware and engineering support from vendors like NVIDIA and AMD.

Overview: AI Code Guide is a comprehensive roadmap designed to help coders integrate AI into their development workflows, whether they're using AI as a co-pilot or aiming to build their own software products with AI assistance. Key features: - Up-to-date information on the latest AI coding tools and practices - Guidance on integrating AI into existing coding workflows - Tips for beginners and experienced developers alike - Resources for staying updated on the rapidly evolving landscape of AI in software development Tech stack: The guide utilizes various programming languages and tools, including Python, JavaScript, and various AI libraries and frameworks. Why it matters: This guide serves as a valuable resource for anyone looking to leverage AI in their coding practices, whether they're just starting out or looking to enhance their existing skills. It provides a clear, accessible overview of the latest trends and practices in AI-assisted coding, making it an essential tool for developers looking to stay ahead in the ever-changing world of software development.

### Overview: 本项目是编程导航AI开发实战的新项目，旨在通过开发AI恋爱大师应用和自主规划智能体YuManus，帮助程序员掌握AI时代的核心技术。项目使用Spring Boot 3、Java 21、Spring AI、RAG知识库、PGVector向量数据库、Tool Calling工具调用等技术，提供从需求分析到项目部署的完整教程，提升求职竞争力。 ### Key features: - **AI恋爱大师应用**：提供情感指导，支持多轮对话、对话记忆、RAG知识库检索、工具调用、MCP服务调用。 - **自主规划智能体YuManus**：基于ReAct模式，可以自主规划约会计划，包括网页搜索、资源下载、PDF生成工具。 - **技术栈**：Java 21 + Spring Boot 3、Spring AI、RAG知识库、PGVector向量数据库、Tool Calling工具调用、MCP模型上下文协议、AI智能体构建、Serverless计算服务、AI大模型开发平台百炼、Cursor AI代码生成、SSE异步推送、第三方接口如SearchAPI、Pexels API、Ollama大模型部署、Kryo高性能序列化、Jsoup网页抓取、iTextPDF生成、Knife4j接口文档。 - **Why it matters**: - 项目新颖，真实业务场景，提升求职竞争力。 - 实战教程，从需求分析到项目部署，全面覆盖程序员必知必会的AI技术。 -

Overview: Mars is a tensor-based unified framework designed for large-scale data computation, scaling numpy, pandas, scikit-learn, and Python functions. Key features: - Tensor-based unified framework - Scalability for numpy, pandas, scikit-learn, and other libraries - Installation via pip or development environment setup Tech stack: Python Why it matters: Mars simplifies complex data computation tasks, making them more efficient and scalable, especially for large datasets. Repository: mars-project/mars GitHub stars: 2744 Primary language: Python Documentation: https://mars-project.readthedocs.io/en/latest/

### Overview: The Open Source Face Recognition SDK is a comprehensive, privacy-focused solution for integrating face recognition capabilities into applications. Built with state-of-the-art deep learning models, it ensures high accuracy and complete data privacy through on-premise processing. ### Key Features: - 🔒 **100% On-Premise**: All processing happens locally, ensuring no data leaves your device. - 🎯 **High Accuracy**: Powered by advanced deep learning models. - ⚡ **Real-Time Processing**: Fast face detection and recognition capabilities. - 🔧 **Easy Integration**: Simple Python APIs for seamless development. - 🌐 **Cross-Platform**: Compatible with Windows and Linux systems. - 📱 **GPU Optional**: Works efficiently on CPU-only systems. - 🆓 **Completely Free**: Open source with no licensing fees. ### Why It Matters: - **Privacy Focus**: Ensures complete data privacy through on-premise processing. - **Accuracy**: State-of-the-art models guarantee high accuracy in face detection and recognition. - **Ease of Use**: Simple Python APIs make integration straightforward. - **Versatility**: Compatible with Windows and Linux, supporting various image formats. ### Tech Stack: - Python ### GitHub Stars: 2739 ### GitHub Description: The world's first completely free and open-source face recognition SDK for Windows and Linux from [Faceplugin](https://faceplugin.com/). This SDK provides high-accuracy face detection and recognition while ensuring complete data privacy through on-premise processing. --- ## 🚀 Overview The **Open Source Face Recognition SDK

### Overview: Webcmd is a self-learning agent browser that helps reduce the cost and time agents spend on website navigation by leveraging the context and memory of real browsers. ### Key features: - **Live Browser Control**: Inspect, click, type, and capture network calls in a real browser. - **Sitemap Memory**: Capture an agent-facing sitemap of observed pages, states, actions, workflows, APIs, pitfalls, and fallback paths. ### Tech stack: TypeScript ### Why it matters: Webcmd simplifies and speeds up website navigation for AI agents, reducing the need for repeated site visits and improving the efficiency and reliability of browser automation.

Overview: KubeRay is a powerful, open-source Kubernetes operator designed to simplify the deployment and management of Ray applications. It offers a suite of custom resource definitions and a dashboard for easy management. Key features: - RayCluster: Fully manages the lifecycle of RayCluster, including cluster creation/deletion, autoscaling, and fault tolerance. - RayJob: Automatically creates a RayCluster and submits a job when the cluster is ready. - RayService: Consists of a RayCluster and Ray Serve deployment graph, offering zero-downtime upgrades and high availability. Tech stack: Go Why it matters: KubeRay simplifies the deployment and management of Ray applications on Kubernetes, making it easier for developers to run Ray on Kubernetes.

### Overview: A text-to-image playground utilizing Stable Diffusion (SD) V2 for generating images from text prompts. The project aims to provide a user-friendly interface for experimenting with image generation using Stable Diffusion, replacing the older DALL-E Mini. ### Key Features: - **Stable Diffusion V2 Support**: Utilizing the latest Stable Diffusion model for generating high-quality images. - **Backend Implementation**: The project is hosted on Google Colab, allowing for real-time image generation and backend management. - **User-Friendly Interface**: A simple GitHub-hosted frontend for easy access and interaction with the backend. ### Tech Stack: - JavaScript - Python - JAX (for GPU support) - Stable Diffusion V2 ### Why It Matters: This project offers a convenient and efficient way to experiment with image generation using Stable Diffusion, making it accessible to a broader audience of developers and enthusiasts.

Overview: Distributed Machine Learning Toolkit (DMTK) is a comprehensive framework for distributed machine learning, including projects like Multiverso, LightLDA, LightGBM, and Distributed Word Embedding. Key features: - Multiverso: Parameter server framework for distributed machine learning. - LightLDA: Scalable, fast, and lightweight system for large-scale topic modeling. - LightGBM: Fast, distributed, high performance gradient boosting framework. - Distributed word embedding: Distributed algorithm for word embedding. Tech stack: C++ and Python/Lua bindings. Why it matters: DMTK is a crucial tool for researchers and developers working on distributed machine learning, providing a robust framework for efficient and scalable training of large models.

Overview: SimpleCV is an open-source framework for machine vision, providing a concise and readable interface for basic and advanced computer vision tasks using Python and OpenCV. Key features: - Easy-to-use camera and image manipulation interface - Interoperable with various image formats and sources - Comprehensive feature extraction and manipulation capabilities - Fast and easy-to-understand linear algebra operations Tech stack: Python and OpenCV Why it matters: SimpleCV is an excellent tool for both beginners and advanced users looking to quickly and easily perform basic to advanced machine vision tasks. Its user-friendly interface and compatibility with a wide range of image sources make it a valuable resource for anyone working in the field of computer vision.

### Overview: AIMET is a software toolkit designed to improve the runtime performance of deep learning models by reducing compute load and memory footprint. It quantizes models using advanced techniques to facilitate deployment on edge devices like mobile phones or laptops. ### Key Features: - AIMET supports models from the ONNX and PyTorch frameworks. - AIMET employs post-training and fine-tuning techniques to minimize accuracy loss during quantization and compression. - AIMET improves the runtime performance of deep learning models by reducing compute load and memory footprint. - AIMET quantizes models using advanced techniques like Data-Free Quantization, providing state-of-the-art INT8 results on several popular models. - AIMET automates optimization of neural networks, avoiding time-consuming and tedious manual tweaking. ### Tech Stack: Python ### Why It Matters: AIMET is designed to work with PyTorch and ONNX models, improving the runtime performance of deep learning models by reducing compute load and memory footprint. It quantizes models using advanced techniques to facilitate deployment on edge devices like mobile phones or laptops by reducing memory footprint. AIMET supports models from the ONNX and PyTorch frameworks, and it employs post-training and fine-tuning techniques to minimize accuracy loss during quantization and compression. It also automates optimization of neural networks, avoiding time-consuming and tedious manual tweaking.

Overview: CV-CUDA is an open-source library designed for GPU-accelerated image processing and computer vision tasks, optimized for cloud-scale applications on NVIDIA platforms. Key features: - GPU-accelerated image resizing - High-throughput, low-latency processing - Built for performance and seamless integration with C/C++ and Python frameworks Tech stack: C++ Why it matters: CV-CUDA offers optimized GPU-accelerated image processing for AI pipelines on NVIDIA cloud, desktop, and edge platforms, enhancing performance and scalability.

### Overview: xalpha 是一个用于基金投资管理的回测引擎，支持从股票信息获取到净值计算，再到账户记录整合和可视化分析的一站式解决方案。它特别适合资金频繁进出的定投型和网格型投资策略的概览与管理。 ### Key features: - **基金股票信息与净值获取**：精确到分的投资账户记录整合分析与丰富可视化。 - **策略回测**：简单的策略回测功能。 - **定时投资提醒**：根据预设策略的定时投资提醒。 - **AI agent 原生支持**：支持 AI 代理原生功能。 - **数据接口统一**：支持通用日线和实时数据获取器，统一接口一行可以获得几乎任何市场上存在的产品价格数据。 - **持仓配置与股票透视**：支持持仓基金组合的底层持仓配置和股票细节透视，掌握底层持仓和跟踪机构股票池与买卖特点。 - **历史数据获取**：支持获取各种金融产品的历史日线数据或实时数据。 - **估值分析**：支持指数、行业、基金和个股的历史估值和即时估值分析。 - **定价可转债**：支持可转债定价。 - **净值预测**：支持核心 QDII 基金的 T-1 净值预测及基于期货的 T-0 实时净值预测。 ### Tech stack: Python ### Why it matters: xalpha 是一个全面的基金投资管理工具，适用于需要精确管理和分析的投资者。它提供了从数据获取