Projects

### Overview: OpenCove is an infinite canvas workspace designed for AI developers, allowing agents, tasks, and notes to be arranged spatially instead of buried in linear chat histories, making workflows easier to reason about and more efficient. ### Key features: - Infinite spatial canvas - Built for CLI agents (Claude Code, Codex) - Context stays visible - Persistent workspaces - Workspace isolation - Rich media and smart layouts - Global search and control center - Parallel execution and comparison of multiple agents ### Tech stack: TypeScript ### Why it matters: OpenCove simplifies complex AI workflows by providing a spatial workspace that enhances context visibility, parallel execution, and efficient project management, making it easier to reason about and work with AI agents and tasks.

### Overview: Middleware is an open-source tool designed to help engineering leaders measure and analyze the effectiveness of their teams using the DORA metrics. The DORA metrics are a set of four key values that provide insights into software delivery performance and operational efficiency. ### Key Features: - Integration with various CI/CD tools - Automated collection and analysis of DORA metrics - Visualization of key performance indicators - Customizable reports and dashboards - Integration with popular project management platforms ### Tech Stack: TypeScript ### Why It Matters: Middleware helps engineering teams measure and analyze their software delivery performance, enabling better decision-making and continuous improvement.

### Overview: This repository provides an advanced Retrieval-Augmented Generation (RAG) solution designed to tackle complex questions that simple semantic similarity-based retrieval cannot solve. It uses sophisticated graph-based algorithms to handle tasks effectively. ### Key features: - Graph-based algorithm for complex question answering - Advanced RAG solution for non-trivial questions - Demonstrates a deterministic graph acting as the "brain" of a controllable autonomous agent ### Tech stack: Jupyter Notebook ### Why it matters: This project showcases a sophisticated deterministic graph acting as the "brain" of a highly controllable autonomous agent capable of answering non-trivial questions from your own data.

### Overview: Imodels is a Python package designed for concise, transparent, and accurate predictive modeling. It provides a simple interface for fitting and using state-of-the-art interpretable models, all compatible with scikit-learn. These models can replace complex black-box models with simpler ones, improving interpretability and computational efficiency without sacrificing predictive accuracy. ### Key Features: - **RuleFit Rule Set**: A rule-based model for classification and regression, providing interpretable predictions. - **HSTree Classifier**: A hierarchical-shrinkage tree classifier for complex data analysis. - **Interpretable Models**: Models that are easy to understand and explain, making them suitable for applications where interpretability is crucial. ### Tech Stack: Imodels is built on top of scikit-learn and uses Jupyter Notebook for interactive and seamless experimentation. ### Why It Matters: Imodels simplifies the process of creating interpretable models, making them accessible to a wide range of users, including data scientists, researchers, and anyone interested in improving model interpretability. By providing a user-friendly interface and compatibility with scikit-learn, Imodels democratizes the use of advanced machine learning techniques, enabling more transparent and understandable models in various applications.

### Overview: FantasyTalking is a state-of-the-art project that generates realistic talking portraits using coherent motion synthesis. This innovative approach allows for the creation of lifelike animations that closely mimic human speech and facial expressions. ### Key Features: 1. **Realistic Animation Generation**: FantasyTalking produces detailed and lifelike animations that closely resemble human speech and facial expressions. 2. **Coherent Motion Synthesis**: The project utilizes advanced techniques to ensure smooth and consistent lip-sync, making the generated portraits appear more natural and engaging. 3. **Model and Inference**: The project includes models and inference scripts for both audio and image-based conditions, enabling flexibility in various applications. ### Tech Stack: - **Python**: The primary programming language used for developing the project. - **Hugging Face**: Utilized for model training and inference, providing pre-trained models and easy access to the Hugging Face Hub. ### Why It Matters: - **State-of-the-Art Performance**: FantasyTalking achieves remarkable results in generating realistic talking portraits, setting new benchmarks in the field. - **Ease of Use**: The project includes detailed documentation and scripts, making it accessible for developers and researchers alike. - **Community Contributions**: The project encourages contributions from the open-source community, fostering a collaborative environment that enhances the overall quality and functionality of the project. ### Repository: - **Repository**: Fantasy-AMAP/fantasy-talking - **GitHub Stars**: 1629 - **GitHub Description**: [ACM MM 2025] FantasyTalking: Realistic Talking Portrait Generation via Coherent Motion

### Overview: mllm is a fast and lightweight multimodal language model inference engine designed specifically for mobile and edge devices. It supports various models such as Qwen3, Qwen3-VL, and Qwen3.5 on Jetson Orin with W4A16 and W8A8 serving, and Qwen3-VL-2B on Jetson Orin 32GB with W8A8 serving. The model achieves significant speedups in prefill and decode throughput, particularly on Jetson Orin. ### Key Features: - **Model Support**: Supports Qwen3, Qwen3-VL, and Qwen3.5 on Jetson Orin with W4A16 and W8A8 serving, and Qwen3-VL-2B on Jetson Orin 32GB with W8A8 serving. - **Quantization Paths**: Utilizes compressed tensors (AWQ) and Triton per-token activation quantization for W4A16, and CUTLASS INT8 GEMM for W8A8. - **Speedup Metrics**: Achieves up to 3.12x prefill speedup on AGX Orin 32GB and about 12243 tok/s prefill throughput for Qwen3-VL-2B W8A8. - **Performance Benchmarks**: Measures strong prefill gains over llama.cpp on Jetson Orin, with prefill throughput comparable to llama.cpp. ### Tech Stack: - **Programming Languages**: C++ -

Overview: This project is a BOSS直聘职位数据抓取工具，通过Chrome DevTools Protocol复用真实登录态，绕过字体反爬，输出含明文薪资的JSON/CSV格式职位数据，并提供薪资分布、技能词频和求职材料优化提示词。 Key features: - 明文薪资（API 模式，绕过字体反爬） - 跨轮新增标记（`is_new`）：每条岗位带布尔字段标识「是否为本次新增」 - JSON / CSV 双格式输出 - 详情页 JD 抓取 + 技能分析 - 抓取后生成聚合摘要 + 可复制提示词 - 增量写入（异常退出不丢数据） - 一键环境检查 + 持久隔离 Chrome/Chromium CDP profile Tech stack: Python Why it matters: This tool simplifies the process of scraping BOSS直聘职位数据，避免了传统爬虫可能遇到的复杂性和风险，同时提供了清晰的薪资分布和技能分析，帮助求职者更好地准备面试和优化简历。 Repository: eatmoreduck/boss-zhipin-scraper Primary language: Python GitHub stars: 1628 GitHub description: Boss直聘爬虫 / BOSS直聘职位数据抓取工具，基于 Chrome CDP 协议复用真实登录态，绕过字体反爬，输出明文薪资 JSON/CSV + 薪资技能分析。A Chrome-CDP-based

Overview: Magic123 is an official PyTorch implementation of Magic123, a method for generating high-quality 3D object images from a single 2D image using both 2D and 3D diffusion priors. This project aims to bridge the gap between 2D and 3D image generation, providing a more comprehensive and realistic 3D representation. Key features: - Magic123 is an official PyTorch implementation. - It leverages both 2D and 3D diffusion priors for generating high-quality 3D object images. - The project is built upon Stable-DreamFusion, a code repository for Stable Diffusion. Tech stack: Python, PyTorch, Stable-DreamFusion. Why it matters: Magic123 offers a more comprehensive approach to 3D object image generation, combining the strengths of 2D and 3D diffusion priors to produce more realistic and detailed results compared to existing methods. This makes it particularly useful in applications requiring high-quality 3D representations from a single 2D image, such as in augmented reality, virtual reality, and other visual applications.

### Overview: WikiChat is an improved RAG (Reinforcement Auto-Generated) designed to prevent hallucinations in large language models. It leverages Wikipedia and a 7-stage pipeline to ensure responses are factual, addressing common issues like outdated or niche information. ### Key Features: - **Improved Multilingual Support**: Supports 25 different Wikipedias via web and API. - **Factual Responses**: Ensures responses are based on factual data from Wikipedia. - **7-Stage Pipeline**: Consists of 7 steps that involve multiple LLM calls to prevent hallucinations. ### Tech Stack: - Python ### Why it Matters: WikiChat helps prevent the hallucination of large language models by ensuring responses are based on factual data from Wikipedia, making it more reliable and trustworthy. ### Repository: - **Repository**: stanford-oval/WikiChat - **GitHub Stars**: 1623 ### Installation: WikiChat requires the following system requirements: - Python 3.9 or higher - Docker To install dependencies: ```bash pip install -r requirements.txt ``` Configure the LLM of your choice: ```bash # Option 1 (Default): Use our free rate-limited Wikipedia search API # Option 2: Build your own index # Option 3: Use an Azure AI deployment of the embedding model # Option 4: Upload a Qdrant index to 🤗 Hub ``` ### Run WikiChat in Terminal: ```bash # Option 1: Use our free rate-limited Wikipedia search

### Overview: Exploiting Cloze Questions for Few-Shot Text Classification and Natural Language Inference. This repository contains the code for the papers "Exploiting Cloze Questions for Few-Shot Text Classification and Natural Language Inference" and "It's Not Just Size That Matters: Small Language Models Are Also Few-Shot Learners". The papers introduce pattern-exploiting training (PET), a semi-supervised training procedure that reformulates input examples as cloze-style phrases. In low-resource settings, PET and iPET significantly outperform regular supervised training, various semi-supervised baselines, and even GPT-3 despite requiring 99.9% less parameters. ### Key features: - Pattern-exploiting training (PET) - Iterative variant of PET (iPET) - Supports 13 different tasks: Yelp Reviews, AG's News, Yahoo Questions, MNLI, X-Stance, SuperGLUE tasks - Requires 99.9% less parameters compared to GPT-3 ### Tech stack: Python ### Why it matters: This technique is crucial for building efficient and effective language models in resource-constrained settings, making it a valuable addition to the developer community.

Overview: ktx is an executable context layer designed to enhance the accuracy and context of data and analytics queries by integrating with your company's existing databases, BI tools, and modeling code. It automatically learns from company knowledge, maps the data stack, and builds a semantic layer to improve query performance and accuracy. Key features: - Automatically learns from company knowledge - Maps the data stack - Builds a semantic layer Tech stack: TypeScript Why it matters: ktx simplifies data querying by automating the process, ensuring that agents can accurately and efficiently retrieve the information they need from your warehouse, without the need for manual upkeep or constant updates. This makes it an invaluable tool for general-purpose agents and data-driven teams looking to streamline their workflows and improve the quality of their data-driven decisions.

Overview: Enterprise™ is a non-deterministic, non-statically typed, and Turing-complete programming language designed to create disruptive computer programs. It aims to redefine the landscape of enterprise development by focusing on practical, real-world applications. Key features: - Non-deterministic execution - Dynamic typing - Turing completeness - Quantum computing capabilities - Disruptive libraries - Control structures Tech stack: JavaScript Why it matters: By making Enterprise™ a real thing, recruiters and hiring managers will no longer be able to discuss "enterprise developers" as it will be considered outdated. Articles about the "Best Programming Languages for Enterprise Development" will lose their meaning, as Enterprise™ will be the go-to language for disruptive market solutions. Repository: joaomilho/Enterprise Primary language: JavaScript GitHub stars: 1621 GitHub description: 🦄 The Enterprise™ programming language README excerpt: # Enterprise™ > Extra! Extra! * [Disruption Driven Development - Mastering Enterprise™ – the book is in the making!](#ddd---the-book) <a href="https://www.linkedin.com/search/results/index/?keywords=Enterprise%E2%84%A2%20programming%20language&origin=GLOBAL_SEARCH_HEADER"><img src="https://img.shields.io/badge/linkedin-ready-0077B5.svg" /></a> <img src="https://img.shields.io/badge/build-failing-red.svg" /> <img src="https://img.shields.io/badge/coverage-

### Overview: Beaver is a modular open-source RLHF framework developed by the PKU-Alignment team at Peking University. It aims to provide training data and a reproducible code pipeline for alignment research, particularly constrained alignment LLM research via Safe RLHF methods. ### Key Features: - Support for SFT, RLHF, and Safe RLHF training for popular pre-trained models: LLaMA, OPT, Baichuan, etc. - Provides a large human-labeled dataset (up to 1M pairs) for reproducible RLHF research. - Supports training for Reward Model & Cost Model, and provides pre-trained checkpoints. - Supports customized parameters and datasets for SFT and RLHF. - Provides multi-scale metrics for safety constraints verification, such as BIG-bench, GPT-4 Evaluation. ### Tech Stack: Python ### Why It Matters: Beaver simplifies the process of constrained value alignment in large language models, making it easier to conduct research in this area. Its modular design and comprehensive dataset enable researchers to quickly test and validate their models, leading to more reliable and effective alignment results.

Overview: Actionbook is an AI agent that accesses the sources behind logins and paywalls, allowing you to automate tasks and deliver finished work rather than just answers. Key features: - Access to sources behind logins - Operates websites you already use - Delivers finished work, not just answers Tech stack: Rust Why it matters: Actionbook provides a faster, more accurate, and concurrent way to automate tasks on websites, eliminating the need for slow, brittle, and one-at-a-time approaches.

### Overview: This handbook serves as a comprehensive resource for researchers and practitioners interested in Text-to-SQL (NL2SQL) techniques. It provides a survey of the latest advancements in the field, including detailed research papers, practical guidance, and benchmark evaluations. The handbook is continuously updated to reflect the evolving landscape of Text-to-SQL research. ### Key Features: - Comprehensive survey of Text-to-SQL advancements. - In-depth research papers on various techniques. - Benchmark evaluations of Text-to-SQL methods. - Practical guidance for researchers and practitioners. ### Tech Stack: Python ### Why It Matters: This handbook is crucial for anyone involved in the development and application of Text-to-SQL systems. It helps researchers stay updated on the latest research trends and provides practical insights for practitioners to effectively integrate Text-to-SQL solutions into their work. ### Repository: - **Repository:** HKUSTDial/NL2SQL_Handbook - **Primary Language:** Python - **GitHub Stars:** 1619 ### Tech Stack: Python ### Why It Matters: This handbook is crucial for anyone involved in the development and application of Text-to-SQL systems. It helps researchers stay updated on the latest research trends and provides practical insights for practitioners to effectively integrate Text-to-SQL solutions into their work.

Overview: Herdr UI is a native Rust/GPUI client for Herdr daemons running on macOS, Linux, and Windows. It provides a seamless interface for viewing terminal sessions, workspaces, Git worktrees, and agent activity directly from the local Herdr daemon. Key features: - Native terminal cell painting - Workspaces and worktrees in a single sidebar - Teleport feature for moving worktrees to different hosts - Review and annotation capabilities for agent changes Tech stack: Rust and GPUI Why it matters: Herdr UI simplifies the management of Herdr daemons by providing a native and efficient way to interact with them, enhancing productivity and reducing the need for additional terminal emulators. Repository: penso/herdr-gpui GitHub stars: 1619

Overview: AICON 是一个自然语言驱动的无限画布工作流，允许创作者通过文本、图片和视频节点组织创作过程，提供从剧本拆解到视频合成的一整套能力。 Key features: - 自动剧本拆解 - 生成角色参考图和关键帧 - 创作分镜和过渡视频 - 集成图文和配音功能 - 管理生成历史和参考素材 - 无缝对接 Bilibili 分发 Tech stack: Python Why it matters: AICON 是一个革命性的工具，它通过自然语言助手和无限画布，使 AI 视频创作真正成为可编辑的工作流，让创作者能够从灵感到成片的全过程都变得简单和高效。 Repository: 869413421/ai-moive-studio Primary language: Python GitHub stars: 1618 GitHub description: 自然语言驱动的无限画布工作流 Agent，让 AI 视频创作第一次真正变成可编辑的工作流。

Overview: Agent Apprenticeship is a living ecosystem where AI agents complete tasks through workflow loops, improve through iterative execution, and are evaluated by mentor agents or humans in the loop. It turns completed work into reusable work experience and data to improve future agents. Key features: - Real-world task execution - Reusable agent lessons - Full agent execution traces - Agent work episodes and task rollouts - Agent work experience compilations - Structured experience compilation records Tech stack: Python Why it matters: Agent Apprenticeship creates a compounding exchange of agent work experience, generating reusable learning signals and advancing complex workflows through agent loops. It enables the continuous improvement of agents through the exchange of real-world tasks and their execution outcomes. Repository: ray-r-ren/agent-apprenticeship Primary language: Python GitHub stars: 1617

Overview: Amica is an open-source interface designed for interactive communication with 3D characters, featuring voice synthesis and speech recognition, and an emotion engine for expressive interactions. Key features: - Customizable 3D characters with natural voice chat and vision communication - Emotion engine for expressive interactions - Voice Activity Detection (VAD) for seamless conversations - Speech recognition for natural language understanding - Text-to-Speech engine for voice synthesis - Integration with various AI technologies for enhanced interactions Tech stack: TypeScript Why it matters: Amica simplifies the process of creating interactive, expressive 3D characters that can communicate through voice and vision, making it accessible for developers and creators alike. The integration of cutting-edge AI technologies ensures a seamless and engaging user experience. Repository: semperai/amica Primary language: TypeScript GitHub stars: 1615 GitHub description: Amica is an open source interface for interactive communication with 3D characters with voice synthesis and speech recognition. README excerpt: <p align="center"> <img src="https://amica.arbius.ai/ogp.png" width="600" style="margin-bottom: 0.2;"/> </p> <h2 align="center"><a href="https://amica.arbius.ai">Amica: Your friendly personal AI</a></h2> <h5 align="center"> If you like our project, please give us a star ⭐ on GitHub.</h2> <h5 align="center"> [![twitter](https://img.sh

Overview: This repository contains slides, notebooks, and datasets for the Machine Learning University (MLU) Computer Vision class, designed to help beginners get started with Computer Vision and learn widely used Machine Learning techniques. Key features: - Slides - Notebooks - Datasets Tech stack: Jupyter Notebook Why it matters: This class provides a comprehensive introduction to Computer Vision, covering essential Machine Learning techniques and practical applications, making it a valuable resource for anyone interested in the field. Repository: aws-samples/aws-machine-learning-university-accelerated-cv GitHub stars: 1616 GitHub description: Machine Learning University: Accelerated Computer Vision Class README excerpt:  ## Course Overview There are three lectures and one final project for this class. Lecture 1 | title | studio lab | | :---: | ---: | | Intro to ML | - | | Intro to Computer Vision | - | | Neural Networks | [](https://studiolab.sagemaker.aws/import/github/aws-samples/aws-machine-learning-university-accelerated-cv/blob/main/notebooks/MLA-CV-DAY1-NN.ipynb)| | Convolutional Neural Networks | [](https://studiolab.sagemaker.aws/import/github/aws-s