Projects

### Overview: Zero is an AI coding agent designed for your local terminal. It allows you to inspect repositories, edit files, run commands, use browser/terminal helpers, and maintain durable local sessions based on your chosen model and permission level. ### Key Features: - **Repository Inspection**: Quickly inspect and analyze your project repositories. - **Command Execution**: Execute commands directly from the terminal, including running scripts and handling outputs. - **Output Formats**: Supports various output formats like JSON and stream JSON for better readability and integration. ### Tech Stack: Go ### Why It Matters: Zero is a powerful tool for developers, offering a seamless and efficient way to interact with your codebase and perform tasks directly from your terminal. Whether you need to fix failing tests, run specific commands, or manage your project's dependencies, Zero provides a convenient and powerful solution. ### Repository: Repository: Twigpine/zero GitHub Stars: 1741 GitHub Description: The coding agent that answers to you, your model, your machine, your rules. ### Partners: - **Twigpine**: The creator of Zero, providing a simple and intuitive interface. - **Xiaomi MiMo**: A popular terminal emulator, enhancing the user experience. - **Novita AI**: A leading AI company offering advanced coding solutions. - **Concentrate**: A comprehensive development environment for developers. - **Exa**: A powerful terminal emulator with advanced features. ### License: MIT License ### Additional Languages: - **English** - **Chinese (README_ZH.md)** ### Discord: Join the community on Discord

### Overview: BirdNET Analyzer is a powerful tool for processing and analyzing large audio datasets or single audio files. It's designed to be accessible to scientists without a background in computer science, making it ideal for acoustic analyses and research. ### Key Features: - **Advanced Models**: Utilizes cutting-edge deep learning models for improved accuracy and efficiency. - **Ease of Use**: Provides command-line tools for quick analysis, making it accessible to non-technical users. - **Scalability**: Capable of handling large datasets efficiently, suitable for scientific research and monitoring projects. ### Tech Stack: - Python 3.12 ### Why It Matters: BirdNET Analyzer is the most advanced version of BirdNET for acoustic analyses, ensuring that scientists can run analysis without the need for extensive knowledge in computer science. It supports a wide range of models and interfaces, making it a valuable resource for researchers in the field of avian diversity monitoring. ### Repository: - **Repository**: birdnet-team/BirdNET-Analyzer - **Primary Language**: Python - **GitHub Stars**: 1741 ### Documentation: - **Documentation**: [https://birdnet-team.github.io/BirdNET-Analyzer/](https://birdnet-team.github.io/BirdNET-Analyzer/) ### Download: - **Windows**: [releases page](https://github.com/birdnet-team/BirdNET-Analyzer/releases/latest) - **macOS**: [releases page](https://github.com/birdnet-team/BirdNET-Analyzer/releases/latest) ### Installation: - **Install from PyPI

Overview: PWC-Net is a state-of-the-art deep learning model for optical flow estimation, combining image pyramid, warping, and cost volume techniques to achieve remarkable results. Key features include end-to-end training with a PyTorch implementation that closely matches the Caffe version. The model has been widely adopted and cited in the field of computer vision. Key features: - End-to-end trainable deep neural network architecture - Combines image pyramid, warping, and cost volume techniques - Achieves state-of-the-art results in optical flow estimation Tech stack: Python Why it matters: PWC-Net demonstrates the effectiveness of combining traditional optical flow techniques with modern deep learning methods, providing a powerful tool for applications requiring high accuracy in optical flow estimation.

### Overview: Index-Translate is a comprehensive multilingual, multimodal translation model family that supports text, vision, and audio translations, as well as controlled dubbing and long-document translation. It is built on Qwen3.5 and includes 150 languages, making it highly versatile for various translation tasks. ### Key Features: - **Index-Translate (Text & Multimodal Vision)**: Covers 150 languages for text translation, including structured content, hard/soft instruction following, and community expressions. - **Index-Echo (Multimodal Audio & Speech)**: End-to-end speech-to-subtitles (S2TT) and cross-lingual voice-cloned dubbing (S2ST), preserving the source speaker's voice characteristics. - **Index-Homura (Controlled Dubbing)**: Dynamically adjusts translations to match a specified target syllable budget for syllable-level timing alignment. - **Index-NativeLong (Long Document)**: Supports long-context input for complete documents, maintaining term and character name consistency across paragraphs. ### Tech Stack: Python ### Why it Matters: Index-Translate is a powerful tool for multilingual and multimodal translation, making it essential for a wide range of applications such as language education, content localization, and multimedia content translation. Its ability to handle complex instructions and maintain consistency across different languages and contexts makes it highly valuable in various translation tasks.

Overview: This repository contains models for semantic segmentation implemented in PyTorch, including FCN, U-Net, SegNet, PSPNet, GCN, and DUC/HDC. It also includes requirements and preparation instructions for running the models. Key features: - Models for semantic segmentation - Implementation in PyTorch - Pretrained models available - Detailed README and configuration files Tech stack: PyTorch Why it matters: This repository provides a comprehensive set of models and tools for semantic segmentation tasks, making it a valuable resource for developers working in computer vision and image processing. Repository: zijundeng/pytorch-semantic-segmentation GitHub stars: 1735 Primary language: Python GitHub description: PyTorch for Semantic Segmentation

### Overview: SmartGPT is an experimental program designed to enhance the capabilities of Large Language Models (LLMs) by breaking down complex tasks into smaller, manageable problems and utilizing external resources to gather information. ### Key Features: - **Modularity**: SmartGPT supports first-class plugin support, allowing for easy customization and composition of Autos. - **Flexibility**: A single `config.yml` file is used to configure all aspects of the program, ensuring consistency and ease of use. - **Consistency**: SmartGPT dynamically executes actions and statically chains tools to provide consistent results. ### Tech Stack: Rust ### Why It Matters: SmartGPT aims to unlock the full potential of LLMs by providing a modular, flexible, and consistent environment for executing complex tasks. Despite its experimental nature, it is currently at the forefront of innovative ideas in the AutoGPT space, though it faces challenges in ecosystem integration and memory management. The project's contributions to the AutoGPT community are significant, even if the stability is not as polished as some other projects.

Overview: 《机器学习理论导引》（宝箱书）的伴读笔记，旨在帮助读者更快理解机器学习理论。 Key features: - 提供概念解释 - 解释证明思路 - 增加实例辅助理解 Tech stack: Jupyter Notebook Why it matters: 《钥匙书》为正在研读《机器学习理论导引》的读者提供了一种辅助理解的工具，帮助他们更快地掌握机器学习理论的核心概念。 Repository: datawhalechina/key-book Primary language: Jupyter Notebook GitHub stars: 1734 GitHub description: 《机器学习理论导引》的伴读笔记，概念解释、证明补充、案例分享。 README excerpt: <div align="center"> <img width="900" height="1600" alt="钥匙书封面" src="./docs/images/preface.jpg" /> # 钥匙书 Key-book 🔑 <p align="center"><em>《机器学习理论导引》伴读笔记：概念解释 · 证明补充 · 案例分享</em></p> <p align="center"> 📖 <a href="https://datawhalechina.github.io/key-book/">在线阅读</a> | 📥 <a href="https://github.com/datawhalechina/key-book/releases/download/preview/key_book.pdf">PDF 下载</a> | 💬 <a href="#交流群">加入交流群</a> | 🍉 <a href="

### Overview: Splash is a local inference engine for Apple silicon, designed to run OpenAI or Anthropic compatible applications. It leverages DFlash 2 speculative decoding, specialized Metal kernels, and automatic memory planning to efficiently handle requests. ### Key features: - **Model Compatibility**: Supports OpenAI and Anthropic models, with a specific draft selected for each model. - **Automatic Memory Planning**: Reuses cached prefixes and batches concurrent requests automatically. - **Quick Start**: Requires Apple M3 or newer, macOS 26.4 or later, and Homebrew. The first run downloads and starts the model, with subsequent starts using cached downloads. - **API Support**: Handles OpenAI Chat Completions, Responses, and Messages, including streaming, tool calls, JSON Schema output, and base64 images and PDFs. - **Model Families**: Supports Qwen3.8-27B and Qwen3.6-35B-A3B, with matching GGUF examples and MLX 4-bit targets. - **Model Loading**: Supports various GGUF examples, including mixed-precision UD formats and vision models. ### Tech stack: C++ - **DFlash 2**: Speculative decoding technology. - **Metal Kernels**: Specialized Metal kernels for efficient processing. - **Automatic Memory Planning**: Efficiently plans and manages memory usage. ### Why it matters: Splash simplifies the process of running OpenAI or Anthropic compatible applications on Apple silicon, making it accessible and efficient for developers and users. Its automatic memory planning and model compatibility ensure smooth operation and

### Overview: LLM for Unity enables seamless integration of Large Language Models (LLMs) within the Unity engine, allowing developers to create intelligent AI characters that players can interact with for an immersive experience. ### Key Features: - Seamless integration of LLMs into Unity. - Creation of AI characters that can be interacted with. - Semantic search system using a Retrieval-Augmented Generation (RAG) system. - Standalone C++/C# library for the LLM backend, [LlamaLib](https://github.com/undreamai/LlamaLib). ### Tech Stack: C# ### Why it Matters: The ability to create intelligent AI characters in Unity with LLMs provides a unique and immersive experience for players, enhancing the game's interaction and story-telling capabilities.

### Overview: This project-based course is designed to teach you how to quickly harness the power of the LangChain library for LLM applications. Build 7 end-to-end working LangChain-based generative AI applications with no fluff, no toy examples — just real projects using real APIs and real-world skills. ### Key Features: - **LangChain Hello World**: Your first AI agent - basic structure and LLM integration. - **Modern Search Agent**: Build search agents using LangChain v.1's `create_agent` interface with custom tools, Tavily integration, and structured outputs. - **Agents Under The Hood**: Understand reasoning and acting patterns in AI agents. - **RAG Gist**: The gist of retrieval-augmented generation. - **System One Models**: Fast, calibrated harness decisions with Jev: model routing and auto-mode tool gating middleware. - **Documentation Helper**: Intelligent documentation assistant. - **Code Interpreter**: AI-powered code execution and analysis. - **Reflection Agent**: Self-improving agent with reflection and critique capabilities. - **Agentic RAG**: Advanced retrieval-augmented generation system. ### Tech Stack: LangChain v0.3+, LangGraph, Pinecone, FAISS, Streamlit ### Why It Matters: Learn RAG, vector databases, prompt engineering, and agent workflows. Follow commits chronologically for step-by-step learning. Build agents that solve actual problems with live APIs, making real-world applications practical and relevant.

### Overview: OpenDataCam is an open-source tool designed to quantify the world by understanding and quantifying moving objects using computer vision. It allows anyone to count moving objects from cameras and videos, making it useful for various use cases such as traffic studies and more. ### Key Features: - **Multiple Object Classes:** OpenDataCam can detect and count a wide range of common objects. - **Fine Grained Counter Logic:** It provides precise and detailed counting capabilities. - **Trajectory Analysis:** Tracks the movement of objects across the scene. - **Real-Time or Pre-Recorded Video Sources:** Supports both live and recorded video feeds. - **Run on Small Devices:** Compatible with small devices and can be deployed in the field or in data centers. - **Data Ownership:** Users retain ownership of the data collected. - **Easy to Use API:** Provides an intuitive API for integrating OpenDataCam into existing systems. ### Tech Stack: OpenDataCam leverages machine learning to detect objects in videos and camera feeds, following the objects as they move across the scene. It is built using JavaScript and Docker for easy deployment. ### Why It Matters: OpenDataCam is a valuable tool for anyone looking to quantify the world, whether for traffic studies, environmental monitoring, or other applications where object counting is essential. Its flexibility and ease of use make it accessible to a wide range of users, from researchers to urban planners.

### Overview: 本项目是一个自然语言处理实验，涉及TF-IDF、文本分类、聚类、词向量、情感识别和关系抽取等技术。通过使用Sougou数据集，我们展示了这些技术的实际应用。 ### Key features: - TF-IDF词频-逆文档频率计算 - 文本分类和聚类 - 词向量表示 - 情感识别模型 - 关系抽取算法 ### Tech stack: Python ### Why it matters: 本项目为自然语言处理领域的初学者提供了一个实用的实验环境，展示了多种NLP技术的实际应用，有助于理解和掌握这些技术的原理和实现方法。

### Overview: This repository contains the implementation of YOLOv2 in Keras with TensorFlow backend, supporting training on custom datasets using various backends like MobileNet and SqueezeNet. The demo application for raccoon detection is accessible via a browser and can be run entirely on Windows with the DeepLearn.js and MobileNet backend. The repository includes a todo list for features such as warmup training, raccoon detection, self-driving car, kangaroo detection, support for multiple GPUs, and mAP evaluation. ### Key features: - YOLOv2 implementation in Keras - Support for MobileNet, SqueezeNet, InceptionV3, and ResNet50 backends - Training with custom datasets - Browser-based raccoon detection demo accessible at https://git.io/vF7vI - Todo list includes warmup training, raccoon detection, self-driving car, kangaroo detection, support for multiple GPUs, and mAP evaluation ### Tech stack: Keras, TensorFlow - Keras: Deep learning framework - TensorFlow: Machine learning library ### Why it matters: This project provides a practical tool for developers and researchers to experiment with YOLOv2, making it easier to train and deploy object detection models on various hardware configurations.

Overview: Agent Flow provides a real-time visualization of Claude Code and Codex agent orchestration, allowing developers to understand agent behavior, debug tool call chains, and track time spent on operations. Key features: - Live agent visualization - Auto-detects Claude Code and Codex sessions - Claude Code hooks for real-time streaming - Codex rollout tailing from Codex's event stream - Multi-session support - Interactive canvas - Timeline & transcript panels - JSONL log file support Tech stack: TypeScript Why it matters: Agent Flow makes Claude Code and Codex more understandable and debuggable, providing insights into agent behavior, tool calls, and execution time, ultimately improving the development experience and game creation process.

### Overview: BrickGPT is an innovative project that leverages text-to-image capabilities to generate physically stable and buildable toy brick models. The repository, authored by a team of researchers at Carnegie Mellon University, showcases a groundbreaking approach to brick modeling that has been recognized with the prestigious ICCV 2025 Best Paper Award. ### Key Features: - **Text-to-Image Generation:** BrickGPT converts textual descriptions into physically stable and buildable brick structures. - **Stability Analysis:** Utilizes advanced physics-based stability analysis to ensure the generated models are physically feasible. - **Interactive Demo:** A Gradio demo is provided for real-time visualization and manipulation of generated brick structures. ### Tech Stack: - Python ### Why It Matters: BrickGPT represents a significant advancement in the field of digital fabrication and educational robotics, offering a practical tool for creating stable and buildable brick models from textual descriptions. Its ability to generate physically stable structures has the potential to revolutionize the way brick models are created and used in educational and creative contexts. --- Repository: AvaLovelace1/BrickGPT Primary Language: Python GitHub Stars: 1730

### Overview: Aria is an AI-powered research assistant designed to enhance your Zotero experience by integrating cutting-edge language models with Zotero's powerful reference management capabilities. Leveraging the GPT Large Language Models, Aria provides features such as drag-and-drop reference management, autocompletion for creators, tags, and more, all while offering visual analysis and the ability to save chat notes and annotations. ### Key Features: - **Drag-and-Drop Reference Management**: Easily drag Zotero items and collections into Aria for quick access and management. - **Autocompletion**: Enhance your Zotero experience by completing common Zotero fields with suggestions. - **Visual Analysis (GPT-4 Vision)**: Visualize and analyze Zotero items and collections with GPT-4 Vision. - **Save Chat as Notes and Annotations**: Capture and save chat conversations for further reference. ### Tech Stack: JavaScript ### Why It Matters: Aria streamlines your Zotero workflow by integrating advanced AI models, making research and reference management more efficient and enjoyable. Its features are particularly useful for researchers, students, and anyone looking to enhance their reference management experience with cutting-edge technology. ### Repository: Repository: lifan0127/ai-research-assistant GitHub Stars: 1730 GitHub Description: Aria is Your AI Research Assistant Powered by GPT Large Language Models GitHub License: MIT License GitHub Releases: Latest release available for Zotero 6 and 7 ### Installation: - Download the latest release (.xpi file) from GitHub: [Download Link

Overview: This library is an end-to-end toolkit designed for editing and rendering motion of 3D characters using deep learning. It includes fundamental and advanced functions for working with 3D character animation in Pytorch, with a focus on motion retargeting and motion style transfer. Key features: - Motion Retargeting: Enables the transfer of motion from one skeleton to another, useful for creating animations that fit different characters. - Motion Style Transfer: Allows the transformation of motion from one video to another, useful for generating new animations based on existing videos. Tech stack: The library is built using Python, PyTorch, and Blender. It provides pre-trained models and demo examples to facilitate the editing and rendering of 3D character animations. Why it matters: This library is significant for the field of computer graphics and animation, offering practical tools for creating realistic and dynamic 3D character animations. Its applications range from game development to film production, where precise and customizable animations are crucial. Repository: DeepMotionEditing/deep-motion-editing GitHub stars: 1729 GitHub description: An end-to-end library for editing and rendering motion of 3D characters with deep learning [SIGGRAPH 2020] GitHub URL: https://github.com/DeepMotionEditing/deep-motion-editing

### Overview: Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think ### Key features: - **Representation Alignment (REPA)**: Aligns noisy input states in diffusion models with representations from pretrained visual encoders to improve training efficiency and generation quality. - **Speed Improvement**: REPA significantly speeds up SiT training by 17.5x. - **State-of-the-Art Performance**: Achieves FID=1.42 on ImageNet. ### Tech stack: Python ### Why it matters: REPA simplifies the training process of diffusion models, making them more efficient and effective. This method has the potential to revolutionize the field of generative models by reducing computational costs and improving generation quality.

Overview: This repository contains a curated list of papers on self-supervised learning applied to graph neural networks (GNNs). It categorizes papers by their publication years, aiming to provide a comprehensive overview of the field. Key features: - Papers on self-supervised learning for GNNs - Papers on community detection - Papers on graph representation learning - Papers on graph anomaly detection - Papers on feature imputation in graphs - Papers on graph contrastive learning - Papers on graph augmentation - Papers on knowledge graph entity alignment - Papers on chemical pre-trained models Tech stack: Python Why it matters: This repository serves as a valuable resource for researchers and practitioners interested in the latest advancements in self-supervised learning techniques applied to GNNs. It provides a structured way to stay updated with the field and can be used as a reference for future research or implementation.

Overview: NeuroNER is a Python-based program designed for named-entity recognition using neural networks. It offers easy-to-use and state-of-the-art results, with support for both CPU and GPU configurations. Key features: - Easy-to-use interface - State-of-the-art NER results - Support for both CPU and GPU - Pretrained models available for download - Word embeddings from GloVe and Stanford models - Sample datasets included for quick start Tech stack: Python, TensorFlow, BRAT (optional), Word embeddings (GloVe, Stanford) Why it matters: NeuroNER simplifies the process of named-entity recognition, making it accessible and effective for developers and researchers alike.