Projects

### Overview: Atmosphere is a real-time engine for AI agents on the JVM, providing streaming capabilities over WebSocket, SSE, long-polling, and gRPC, with optional WebTransport/HTTP/3. Tokens flow from the LLM runtime to the client through a broadcaster, allowing for filtered, gated, and observed interactions. It supports governance, human approval, and session replay, making it suitable for teams needing AI agents to behave like production services. ### Key Features: - **Real-Time Streaming:** Tokens flow from the LLM runtime to the client through a broadcaster, allowing for filtered, gated, and observed interactions. - **Governance:** A plain `@Agent` is a full deep agent, batteries-included, with features like policy admission, human approval, and session replay. - **Multi-Stack Support:** Supports multiple AI frameworks, including LangChain4j, Anthropic, and 9 more, with a SPI (Service Provider Interface) for easy integration. - **Reconnect and Authorization:** Handles reconnect, authorization, and governance, ensuring seamless operation across different AI frameworks. - **Session Replay:** An opt-in session tape records session-boundary AI events, allowing for reconstructing and distilling model outputs. ### Tech Stack: Java ### Why It Matters: Atmosphere is built for teams that need AI agents to behave like production services, providing streaming over real transports, guard before every tool call, observable by tenant, and portable across AI frameworks without rewriting the endpoint. It supports governance, human approval, session replay, and seamless integration with various AI frameworks

### Overview: Lightly**SSL** is a Python library designed for self-supervised learning on images. It provides a modular framework that exposes low-level building blocks, making it easy to use and written in a PyTorch-like style. The library supports custom backbone models and distributed training using PyTorch Lightning, making it accessible for both research and production use. ### Key Features: - **Modular Framework**: Exposes low-level building blocks for loss functions and model heads. - **Easy to Use**: Written in a PyTorch-like style, making it straightforward to integrate into existing projects. - **Custom Backbones**: Supports custom backbone models for self-supervised pre-training. - **Distributed Training**: Utilizes PyTorch Lightning for distributed training, enabling efficient use of multiple GPUs. - **Support for Colab**: Provides examples in both PyTorch and PyTorch Lightning for easy experimentation. ### Tech Stack: - Python - PyTorch - PyTorch Lightning ### Why It Matters: Lightly**SSL** simplifies the process of implementing self-supervised learning on images, making it accessible for researchers and developers. The library's modular design and support for distributed training make it suitable for both academic and industrial applications.

### Overview: Qiushi-Skill 是一个集成了经典唯物辩证法与实践哲学方法论的AI技能集合，旨在武装AI的大脑，帮助开发者系统性地分析问题和解决问题。 ### Key features: - 总原则：实事求是 - 矛盾分析法 - 实践认识论 - 调查研究 - 群众路线 - 批评与自我批评 - 持久战略 - 集中兵力 - 星火燎原 - 统筹兼顾 ### Tech stack: JavaScript ### Why it matters: 经典唯物辩证法与实践哲学方法论可以用于指导任何需要分析问题和解决问题的场景，帮助开发者提高决策质量和工作效率。

Overview: DiscoArt is a Python library designed for creating Disco Diffusion artworks, offering a modern and professional API with features like result recovery and persistence, and ease of integration with larger cross-modal applications. Key features: - Create Disco Diffusion artworks in one line - Result recovery and persistence - GRPC/HTTP serving w/o TLS - Post-analysis - Easing the integration to larger cross-modal or multi-modal applications Tech stack: Python Why it matters: DiscoArt simplifies the process of creating Disco Diffusion artworks, making it accessible to both generative artists and AI enthusiasts, while ensuring high usability and maintainability through features like result recovery and persistence. It also integrates seamlessly with other cross-modal applications, making it a versatile tool for developers. Repository: jina-ai/discoart GitHub stars: 3818 GitHub description: 🪩 Create Disco Diffusion artworks in one line

### Overview: LLamaSharp is a C#/.NET library designed to efficiently run large language models (LLMs) like 🦙LLaMA on your local device. It leverages the powerful llama.cpp framework for inference and provides a higher-level API for seamless integration into various applications. ### Key features: - **Cross-Platform Support**: Runs on both CPU and GPU, ensuring compatibility across different environments. - **Efficient Inference**: Optimized for both CPU and GPU, providing fast performance. - **RAG Support**: Enables the use of rich text generation in LLMs. - **Easy Integration**: Provides libraries for integrating with popular AI frameworks like BotSharp, Langchain, and MaIN.NET, making it simple to deploy LLMs in your applications. ### Tech stack: - **C#/.NET**: The primary programming language used for development. - **llama.cpp**: The underlying framework that powers the inference process. ### Why it matters: - **Efficiency**: Runs inference efficiently on both CPU and GPU, making it ideal for applications requiring high performance. - **Ease of Use**: Provides a higher-level API and RAG support, making it straightforward to integrate LLMs into various applications. - **Community Support**: The library is actively maintained and has a growing community, ensuring ongoing support and updates. ### Repository: - **Repository**: SciSharp/LLamaSharp - **GitHub Stars**: 3815 - **GitHub Description**: A C#/.NET library to run LLM (🦙LLaMA/LLaVA) on

### Overview: This repository contains Jupyter notebooks from the scikit-learn video series, providing a comprehensive introduction to Machine Learning using Python's scikit-learn library. The series includes 10 video tutorials and 4.5 hours of content, covering everything from what Machine Learning is and how it works to training models and evaluating performance. ### Key Features: - **Machine Learning Introduction**: What is Machine Learning, its two main categories, and examples. - **Setting Up Python for Machine Learning**: Benefits of scikit-learn, installation, and Jupyter Notebook usage. - **Iris Dataset**: Loading and describing the iris dataset. - **Model Training**: K-nearest neighbors classification model, four steps for model training. - **Model Evaluation**: Choosing models, tuning parameters, and performance estimation. - **Data Science Pipeline**: Using pandas and seaborn for data manipulation and visualization. ### Tech Stack: Python 3.9.1, scikit-learn 0.23.2. ### Why it Matters: This series provides a practical, step-by-step guide to Machine Learning, making complex concepts accessible through clear, interactive Jupyter notebooks. Ideal for beginners and those looking to refresh their skills in Python and Machine Learning.

Overview: Dotted thought-orb loading indicators for AI & agent UIs, featuring nine hand-tuned animated states in two sizes, rendered on a plain 2D canvas without WebGL or filters. Key features: - Nine hand-tuned animated states - Two tuned presets for inline-text and chat-avatar scale - Automatic theme detection from host project - Strictly monochrome with automatic theme switching - Performance optimizations for low-end devices Tech stack: TypeScript Why it matters: These thought-orb indicators provide visually appealing and responsive loading indicators for AI and agent UIs, enhancing the user experience with minimal resources and performance optimizations.

Overview: LazyCodex is a comprehensive agent harness designed specifically for complex codebases, offering project memory, planning, execution, and verified completion within Codex. Key features: - Project memory - Planning - Execution - Verified completion Tech stack: TypeScript Why it matters: LazyCodex streamlines the development process by automating tasks, ensuring project integrity, and providing a seamless workflow for complex codebases, making it an essential tool for developers. Repository: code-yeongyu/lazycodex Primary language: TypeScript GitHub stars: 3804 GitHub description: The one and only agent harness for complex codebases. Project memory, planning, execution, and verified completion inside Codex.

Overview: This book serves as an introduction to probability and statistics, focusing on practical applications and avoiding complex mathematical notation. Key features: - Explains statistical concepts using simple language - Includes numerous examples and exercises - Covers both theoretical and practical aspects of probability and statistics Tech stack: Jupyter Notebook Why it matters: This book is essential for developers looking to understand the basics of probability and statistics without getting bogged down in advanced mathematical concepts. Repository: Visualize-ML/Book5_Essentials-of-Probability-and-Statistics Primary language: Jupyter Notebook GitHub stars: 3803 GitHub description: Book_5_《统计至简》 | 鸢尾花书：从加减乘除到机器学习；上架！ README excerpt: 《统计至简》五折入口： https://zhuanlan.zhihu.com/p/634253719 <br> 《数学要素》五折入口： https://zhuanlan.zhihu.com/p/620243026 <br> 《矩阵力量》五折入口： https://zhuanlan.zhihu.com/p/634253719 看个人情况，开源资源，永久有效哈。 纠错多的同学会得到赠书，以示感谢。

Overview: XIAOJUSURVEY 是一个轻量级、安全的调研系统，提供一站式的问卷、考试、投票、测评和复杂表单解决方案，适用于市场调研、客户满意度调查、在线考试、投票、报道和测评等场景。 Key features: - 高效问卷设计与管理 - 多终端适配与集成 - 数据在线分析与导出 - AI 生成问卷功能 - 精细权限管理与团队协同 - 数据加密传输与敏感信息检测 Tech stack: TypeScript Why it matters: XIAOJUSURVEY 提供了全面的调研解决方案，帮助企业快速构建和管理各类调研活动，同时确保数据安全和高效分析，助力企业提升市场洞察力和客户满意度。

Overview: Knowhere is a document parsing and retrieval system designed to prepare unstructured data for AI agents, especially across local and offline document collections. It ingests unstructured documents and produces persistent, navigable memory, making it a natural fit for *Agentic RAG*, *vector-based RAG*, or any LLM workflow. Key features: - Knowhere ingests and parses unstructured documents. - It reconstructs a hierarchical structure and multi-modal structuring. - Knowhere constructs graphs to represent the data. - It outputs structured chunks ready for AI agents. Tech stack: Python Why it matters: Knowhere simplifies the process of preparing unstructured data for AI agents, making it a valuable tool for developers and researchers looking to integrate AI into their workflows.

Overview: The Google Antigravity SDK is a Python library designed for building AI agents that leverage the full power of Google Antigravity and Gemini. It provides a secure, scalable, and stateful infrastructure layer to abstract the agentic loop, allowing developers to focus on what their agent does rather than how it runs. Key features: - Secure, scalable, and stateful infrastructure layer - Abstracts the agentic loop - Supports both Express Mode (API Key) and Standard Mode (Project & Location with ADC) - Provides examples for quickstart and complete example Tech stack: Python Why it matters: The Google Antigravity SDK simplifies the development of AI agents by abstracting the complexities of running an agent, making it easier to focus on the core functionality of the agent itself. This is particularly useful in enterprise environments where secure and scalable deployment is crucial. Repository: google-antigravity/antigravity-sdk-python GitHub stars: 3797

### Overview: MAGI-1 is a state-of-the-art autoregressive video generation model designed to scale up to high-resolution videos at unprecedented speeds. Leveraging advanced neural architectures and efficient hardware acceleration, MAGI-1 achieves remarkable performance in generating high-quality videos with minimal latency, making it ideal for applications requiring real-time or near-real-time video synthesis. ### Key Features: - **High-Resolution Video Generation**: MAGI-1 can generate videos at resolutions up to 1080p with minimal latency. - **Autoregressive Architecture**: Uses an autoregressive model to generate videos in a single pass, ensuring smooth and continuous motion. - **Efficient Hardware Acceleration**: Utilizes specialized hardware to speed up the generation process, making it suitable for real-time applications. ### Tech Stack: - **Python**: The primary programming language used to develop the model and its accompanying tools. ### Why It Matters: MAGI-1 is crucial for applications requiring high-quality video synthesis, such as video editing, content creation, and interactive media experiences. Its ability to generate videos at scale and in real-time makes it a game-changer in the field of video synthesis and generation. Repository: SandAI-org/MAGI-1 Primary language: Python GitHub stars: 3795

Overview: This project is an AI assistant bot that combines the capabilities of ChatGPT and DALL-E 2 to interact with users through WhatsApp. Key features: - Uses OpenAI's GPT and DALL-E 2 for intelligent responses. - Transcribes voice messages for user input. - Integrates with WhatsApp to send and receive messages. - Requires Node.js, npm, OpenAI API key, and a WhatsApp account. Tech stack: TypeScript Why it matters: This project showcases the integration of cutting-edge AI technologies with traditional communication tools, potentially revolutionizing how users interact with AI assistants. Repository: askrella/whatsapp-chatgpt GitHub stars: 3794

Overview: Streamer-Sales 销冠 —— 卖货主播大模型是一个能够根据商品特点激发用户购买意愿的智能模型，通过生动的解说和多种功能，提升销售效率和用户体验。 Key features: - 一键生成主播文案 - 使用 LMDeploy 加速推理 - 支持 RAG 检索增强生成 - 文字转语音 - 数字人生成 - 语音转文字 - Vue + pinia + element-plus 前端搭建 - FastAPI 后端部署 - Docker-compose 部署 Tech stack: Python Why it matters: 通过结合先进的技术，如 LMDeploy、RAG 检索增强、TTS 文字转语音、数字人生成等，Streamer-Sales 销冠 —— 卖货主播大模型能够提供更加智能和个性化的商品解说，提升销售效果，增强用户购物体验。 Repository: PeterH0323/Streamer-Sales

Overview: OpenClaw 汉化发行版是基于 GitHub 195,000+ Stars 的开源个人 AI 助手平台，提供CLI和Dashboard全中文界面，支持多种聊天应用接入。 Key features: - 每小时自动同步 - 全中文界面 - 支持多种聊天应用接入 - 免费词元 API 赞助推广 Tech stack: JavaScript Why it matters: 本项目为OpenClaw提供全中文界面和CLI/Dashboard控制台，方便用户在各种聊天应用中使用AI助手，同时支持词元API进行免费推广，提升用户体验。

### Overview: 本项目是关于机器学习手推笔记的GitHub仓库，涵盖了从第一章到第十六章的详细内容。手推笔记是根据周志华的《机器学习》一书进行的，笔记包括模型评估与选择、线性模型、决策树、神经网络、支持向量机、贝叶斯分类器、集成信息、聚类、降维与度量学习、特征选择与稀疏学习、计算学习理论、半监督学习、概率图模型、规则学习和强化学习等内容。手推笔记是作者王博Kings在2021年3月13日更新完成的，总共有16章，内容详尽且具有很高的学习价值。 ### Key features: - 手推笔记涵盖了机器学习的多个重要章节，包括模型评估与选择、线性模型、决策树、神经网络、支持向量机、贝叶斯分类器、集成信息、聚类、降维与度量学习、特征选择与稀疏学习、计算学习理论、半监督学习、概率图模型、规则学习和强化学习。 - 手推笔记是根据周志华的《机器学习》一书进行的，内容详细且具有很高的学习价值。 - 手推笔记是作者王博Kings在2021年3月13日更新完成的，总共有16章，内容详尽且具有很高的学习价值。 ### Tech stack: 手推笔记使用了Markdown格式进行编写，包括表格、图片

Overview: Hunyuan-PromptEnhancer is a versatile tool designed to enhance text-to-image and image-to-image prompt rewriting, ensuring clear and structured prompts for better image generation. Key features: - Dual-mode support for Text-to-Image and Image-to-Image editing - Intent preservation of key elements across rewriting - Robust parsing mechanism for reliable output - Flexible deployment options for full-precision, quantized, and vision-language models Tech stack: Python Why it matters: By improving the clarity and structure of prompts, this tool enhances the effectiveness of image generation models, potentially leading to more accurate and visually appealing results.

### Overview: AgentBench is a comprehensive benchmark for evaluating large language models as agents, introduced at ICLR 2024. It includes a function-calling version integrated with AgentRL, supporting various tasks like `alfworld`, `dbbench`, `knowledgegraph`, `os_interaction`, and `webshop`. ### Key Features: - Function-calling prompt style - Fully-containerized deployment support - Integration with AgentRL framework - Support for multiple tasks including `alfworld`, `dbbench`, `knowledgegraph`, `os_interaction`, and `webshop` ### Tech Stack: Python ### Why it Matters: AgentBench provides a standardized benchmark to evaluate the performance of large language models as agents, ensuring fair and comparable results across different models and tasks.

### Overview: TensorFlow 是一个开源的机器学习库，用于构建和部署深度神经网络。它支持多种编程语言，包括 Python 和 C++，并提供稳定和开发版的版本。 ### Key Features: - **Python API**: TensorFlow 的主要 API，支持数据流图进行数值计算。 - **C++ API**: 提供了没有向后兼容性保证的 C++ API。 - **开发版**: TensorFlow 的开发版，提供 pip 包 tf-nightly 和 tf-nightly-gpu。 - **TensorBoard**: 数据可视化工具包，用于数据可视化。 - **安装指南**: 包含稳定二进制版的安装步骤和从源码安装的安装步骤。 - **贡献指南**: 提供了如何参与贡献 TensorFlow 的指南。 - **持续构建状态**: 包含官方构建状态的持续状态和相应的 Artifacts。 ### Tech Stack: - Python - C++ - Java - JavaScript - Go - Swift ### Why It Matters: TensorFlow 是机器学习和深度学习领域的关键工具，广泛应用于图像识别、自然语言处理、语音识别等多个领域。它的灵活性和可扩展性使其成为研究和生产环境中的首选工具。