Projects

Overview: 🐸STT is a fast, open-source, multi-platform deep-learning toolkit designed for training and deploying speech-to-text models. Key features: - High-quality pre-trained STT model - Efficient training pipeline with Multi-GPU support - Streaming inference - Multiple possible transcripts, each with an associated confidence score - Real-time inference - Small-footprint acoustic model - Bindings for various programming languages Tech stack: C++ Why it matters: 🐸STT is battle-tested in both production and research, making it a reliable choice for developers looking to integrate speech-to-text capabilities into their projects.

### Overview: Agentic Context Engine (ACE) is a powerful AI solution that allows agents to learn from experience, making them more efficient and effective. With its hosted solution available at kayba.ai, ACE automates the process of training and deploying agents, ensuring they perform optimally without manual intervention. ### Key Features: - **Automated Learning**: Agents learn from past interactions to improve their performance. - **Scalability**: Easily scalable to handle large volumes of data and complex tasks. - **Ease of Use**: Simple setup and integration with existing workflows. ### Tech Stack: Python ### Why It Matters: ACE is the foundation of Kayba, a platform that automates the entire process of agent development, training, and deployment. By leveraging AI, Kayba helps businesses streamline their operations and reduce the need for manual intervention, leading to significant efficiency gains and cost savings.

### Overview: `open-computer-use` is an open-source project that provides an alternative to Codex Computer Use, offering a non-intrusive way to run Computer Use on macOS, Linux, and Windows. This project is inspired by OpenAI's Codex Computer Use and aims to create an open-source version. ### Key features: - **Computer Use on macOS, Linux, and Windows**: Run Computer Use on these operating systems. - **Swift Tech Stack**: The project is built using Swift, making it compatible with macOS, Linux, and Windows. - **GitHub Stars**: The repository has received 2,603 stars, indicating its popularity and community interest. ### Why it matters: This project provides an alternative to the intrusive nature of traditional Computer Use, allowing for a more seamless and user-friendly experience. It is built on the principles of accessibility and open-source development, making it a valuable tool for developers looking to integrate Computer Use into their projects without compromising user experience. ### Tech stack: Swift The project is primarily developed using Swift, a powerful and efficient programming language, ensuring compatibility across macOS, Linux, and Windows. ### Repository: iFurySt/open-computer-use The primary language used in this project is Swift, and the repository is hosted on GitHub, with 2,603 stars indicating its popularity and community interest.

### Overview: A low-cost motion capture system designed for indoor drone autonomy, enabling multiple drones to navigate and operate efficiently within a room. ### Key features: - **Room-Scale Tracking**: Tracks multiple drones within a room using external cameras. - **Autonomous Flight**: Allows drones to autonomously navigate and perform tasks indoors. - **Open-Source Architecture**: Built from the ground up, using open-source libraries and components. ### Tech stack: TypeScript ### Why it matters: This system simplifies the process of motion capture and drone autonomy, making it accessible for various applications, such as indoor drone delivery, surveillance, and more. Repository: jyjblrd/Low-Cost-Mocap Primary language: TypeScript GitHub stars: 2603 GitHub description: Low cost motion capture system for room scale tracking YouTube Video: [https://youtu.be/0ql20JKrscQ?si=jkxyOe-iCG7fa5th](https://youtu.be/0ql20JKrscQ?si=jkxyOe-iCG7fa5th) Architectural Diagram:  Dependencies: Install the pseyepy python library: [https://github.com/bensondaled/pseyepy](https://github.com/bensondaled/pseyepy) Install npm and yarn Runing the code: From the computer_code directory, run `yarn install` to

Overview: This repository contains the code and slides for 15 homework assignments from the Machine Learning course taught by 李宏毅 (Hung-yi Lee) at National Taiwan University in Spring 2022. Key features: - Homework assignments covering various machine learning techniques such as regression, classification, CNNs, self-attention, transformers, BERT, autoencoders, explainable AI, adversarial attack, reinforcement learning, network compression, life-long learning, and meta learning. - Code and slides are available for each homework assignment. - The repository includes lecture videos from the course. Tech stack: Jupyter Notebook Why it matters: This repository provides a comprehensive set of resources for students and researchers interested in machine learning, allowing them to practice and deepen their understanding of various algorithms and techniques. GitHub stars: 2603 Repository: virginiakm1988/ML2022-Spring

### Overview: MiniMax-M2 is a compact, fast, and cost-effective MoE model designed for elite performance in coding and agentic tasks, all while maintaining powerful general intelligence. It is optimized for elite performance in coding and agentic workflows, making it a valuable tool for developers and AI agents alike. ### Key Features: - **Superior Intelligence**: Benchmarks from Artificial Analysis rank MiniMax-M2 highly in general intelligence across mathematics, science, instruction following, coding, and agentic tool use. - **Advanced Coding**: Excels at multi-file edits, coding-run-fix loops, and test-validated repairs, demonstrating practical effectiveness in terminals, IDEs, and CI across languages. - **Agent Performance**: Plans and executes complex, long-horizon toolchains across shell, browser, retrieval, and code runners, consistently locating hard-to-surface sources, maintaining evidence traceable, and gracefully recovering from flaky steps. - **Efficient Design**: With 10 billion activated parameters (230 billion in total), MiniMax-M2 delivers lower latency, lower cost, and higher throughput for interactive agents and batched sampling, perfectly aligned with the shift toward highly deployable models. ### Tech Stack: Python, PyTorch, TensorFlow, and Hugging Face Transformers. ### Why It Matters: MiniMax-M2 is a game-changer for developers and AI agents, offering superior intelligence, advanced coding capabilities, and efficient design, making it a valuable tool for anyone looking to improve their coding and agentic workflows.

Overview: APKiD is a tool designed to identify various types of compilers, packers, and other obfuscation techniques used in Android APKs. It provides a comprehensive fingerprinting of Android applications, similar to the PEiD tool for Windows executables. Key features: - Identifies compilers, packers, and obfuscators - Scans for various Android application obfuscation methods - Supports multiple file types including APKs and .dex files - Can be used with Docker for easy integration Tech stack: Python Why it matters: APKiD offers a powerful way to analyze and fingerprint Android applications, helping developers identify and mitigate potential security threats and obfuscation techniques used by malicious or pirated apps. Repository: rednaga/APKiD GitHub stars: 2601 GitHub description: Android Application Identifier for Packers, Protectors, Obfuscators and Oddities - PEiD for Android

Overview: Asteroid is a PyTorch-based audio source separation toolkit designed for researchers, offering a comprehensive set of features and recipes to facilitate experimentation on common datasets. Key features: - Fast experimentation on various datasets - Support for a wide range of datasets and architectures - Recipes for reproducing important papers in the field - Pretrained models for quick setup Tech stack: Python Why it matters: Asteroid provides a powerful and flexible solution for audio source separation research, making it an essential tool for researchers and developers in the field. Repository: asteroid-team/asteroid Primary language: Python GitHub stars: 2597 GitHub description: The PyTorch-based audio source separation toolkit for researchers README excerpt: [Repository description](https://github.com/asteroid-team/asteroid)

Overview: Cradle is a framework designed to empower foundation models to perform complex computer tasks by providing a unified interface similar to human interactions, enabling strong reasoning abilities, self-improvement, and skill curation within a standardized general environment. Key features: - Support for multiple games and software applications - Extensive updates and improvements - Integration with OpenAI and Claude APIs Tech stack: Python Why it matters: Cradle simplifies the process of using foundation models for complex tasks by providing a user-friendly interface and enabling seamless integration with various software and APIs. Repository: BAAI-Agents/Cradle GitHub stars: 2594 GitHub description: The Cradle framework is a first attempt at General Computer Control (GCC). It supports agents to ace any computer task by enabling strong reasoning abilities, self-improvment, and skill curation, in a standardized general environment with minimal requirements.

Overview: Pipcook is a machine learning platform designed for JavaScript developers, enabling them to harness the power of machine learning without prior knowledge. Key features: - Pipcook Pipeline: A framework for building and managing machine learning pipelines. - Pipcook Core: A JavaScript runtime for machine learning applications. - Pipcook CLI: A command-line interface for quick experimentation and deployment. Tech stack: TypeScript Why it matters: Pipcook simplifies the process of integrating machine learning into JavaScript applications, making it accessible to developers without extensive knowledge in machine learning or deep learning. Repository: alibaba/pipcook Primary language: TypeScript GitHub stars: 2593 GitHub description: Machine learning platform for Web developers README excerpt: A JavaScript application framework for machine learning and its engineering. Documentation: English | Chinese

Overview: LibMTL is an open-source library built on PyTorch for Multi-Task Learning. It provides a comprehensive set of tools and techniques for solving multi-task learning problems. Key features: - Support for various bilevel methods, including Auto-Lambda, FAMO, MoDo, MOML, FORUM, and STCH. - Integration with PyTorch for efficient model training and inference. - Documentation available in the latest version on ReadTheDocs. Tech stack: PyTorch. Why it matters: LibMTL offers a powerful and flexible framework for researchers and practitioners working on multi-task learning problems, making it an essential tool for anyone interested in this field.

### Overview: Scanpy is a Python library designed for the analysis of single-cell gene expression data. It provides tools for preprocessing, visualization, clustering, trajectory inference, and differential expression testing, making it suitable for datasets ranging from small to large, including those with more than one million cells. The library is compatible with the `anndata` library and supports the `dask` library for handling large datasets, making it efficient even for datasets that exceed the memory capacity of standard systems. ### Key features: - **Scalability**: Designed to handle datasets of up to 100 million cells, making it suitable for large-scale single-cell analysis. - **Python-based Implementation**: Efficiently processes datasets using Python, ensuring compatibility with various Python environments. - **Visualization**: Provides tools for visualizing single-cell data, aiding in the interpretation of complex datasets. - **Clustering and Trajectory Inference**: Facilitates the identification of cell types and their developmental trajectories. - **Differential Expression Testing**: Enables the analysis of gene expression differences between different cell populations. ### Tech stack: - Python ### Why it matters: Scanpy is essential for researchers and data analysts working with large-scale single-cell data, providing a comprehensive toolkit that simplifies the process of analyzing and interpreting single-cell gene expression data. Its ability to handle datasets of up to 100 million cells and its compatibility with `dask` make it a valuable tool for researchers in the field of single-cell biology. ### Repository: - **Repository**: `scverse/scanpy` - **GitHub Stars**: 2

### Overview: Lealone 是一个高性能的全栈自进化通用智能体，能够彻底颠覆现有的应用软件开发模式，适用于氛围编程和各种规模的企业应用。 ### Key features: - 全栈自进化通用智能体 - 安全适用于氛围编程和企业应用 - 高性能的开发工具和运行环境 ### Tech stack: Java ### Why it matters: Lealone 是一个革命性的开发工具，能够大幅度提高开发效率，降低开发成本，适用于各种规模的企业应用，是企业级开发的首选工具。 ### Repository: lealone/Lealone ### GitHub stars: 2587 ### GitHub description: 能安全适用于氛围编程和企业应用的全栈自进化通用智能体

### Overview: The DIS project aims to develop highly accurate dichotomous image segmentation, specifically addressing the challenges in detecting and segmenting objects and backgrounds in images. This project is part of the ECCV 2022 conference and has been widely recognized, with a significant number of GitHub stars indicating its importance and interest. ### Key Features: - **DIS Dataset V1.0 (DIS5K):** A dataset containing images annotated for dichotomous segmentation, focusing on categories such as animals, humans, and cars. - **DIS Dataset V2.0:** An expanded dataset with more than 200 categories, aiming to improve the robustness and accuracy of the models. - **Model Optimization:** The project includes an optimized model for general use, which is different from the academic version used in the DIS V1.0 dataset. ### Tech Stack: The project is primarily developed using Jupyter Notebook, leveraging Python and other relevant technologies for image processing and machine learning. ### Why It Matters: The DIS project is crucial for advancing the field of image segmentation, particularly in detecting and segmenting objects and backgrounds in images. Its comprehensive and large-scale nature ensures better performance across various real-world scenarios, making it valuable for both academic research and practical applications. ### Repository: The project is hosted on GitHub at [xuebinqin/DIS](https://github.com/xuebinqin/DIS). It has received 2,585 GitHub stars, indicating high interest and engagement from the developer community. ### Updates: - **DIS Demo on Bohrium Notebook

Overview: x64dbg-MCP Server is a native MCP plugin for x64dbg that allows users to control the debugger programmatically via HTTP, enabling features such as setting breakpoints, stepping through code, reading memory, and more. It supports zero dependencies and cross-compiles to both x32 and x64. Key features: - Full debugger control with 84 MCP Tools - 22 Event Callbacks for full debugger event coverage - Zero dependencies, single-binary output, and cross-compiles to both x32 and x64 - Streamable HTTP + SSE transport - Bearer Auth with auto-generated token on first run - Config dialog for changing settings - Auto-start on x64dbg launch - Cross-compile from Linux, macOS, or WSL Tech stack: Zig Why it matters: x64dbg-MCP Server provides a powerful and flexible way to interact with x64dbg programmatically, enhancing its capabilities and making it more accessible for reverse engineering and debugging tasks.

### Overview: Griptape is a Python framework designed to simplify the development of generative AI (genAI) applications. It offers a set of straightforward, flexible abstractions for working with large language models (LLMs), retrieval-augmented generation (RAG), and more. ### Key Features: - **Agents**: Consist of a single Task, configured for Agent-specific behavior. - **Pipelines**: Organize a sequence of Tasks to ensure smooth flow of output. - **Workflows**: Configure Tasks to operate in parallel. - **Tasks**: Core building blocks enabling interaction with Engines, Tools, and other Griptape components. - **Memory**: Conversation Memory for LLMs, Task Memory for sensitive outputs, and Meta Memory for additional metadata. - **Drivers**: Facilitate interactions with external resources and services, allowing minimal changes to business logic. - **LLM & Orchestration**: Prompt Drivers for LLM interactions, Assistant Drivers for various services, Ruleset Drivers for external rulesets, Conversation Memory Drivers for conversational data, Event Listener Drivers for external services, Structure Run Drivers for local or cloud execution. - **Retrieval & Storage**: Embedding Drivers for vector embeddings, Rerank Drivers for improved relevance, Vector Store Drivers for storage and retrieval of embeddings, File Manager Drivers for file operations, SQL Drivers for database interactions. - **Multimodal**: Image Generation Drivers, Text to Speech Drivers, Audio Transcription Drivers. - **Web**: Web Search Drivers for web information, Web Scraper Drivers for data extraction. - **Observability**:

### Overview: 本项目旨在收集和梳理垂直领域的开源模型、数据集及评测基准，涵盖医疗、法律、金融、教育等多个领域。 ### Key features: - 模型 - LLaMA2 - Qwen-14B - Qwen-7B - InternLM - Zhongjing - WeaverBird - 夫子•明察 - 数据集 - 评测基准 ### Tech stack: Unknown ### Why it matters: 推动大语言模型在垂直领域的应用，赋能各行各业。

Overview: Beacon is an open-source memory layer designed for AI coding agents, capturing session history across multiple harnesses and turning useful workflows, corrections, and debugging patterns into reusable knowledge for future agents. Key features: - Cross-harness history - Knowledge that compounds - Shared agent memory - Exact session replay - Local-first portability Tech stack: Go Why it matters: Beacon simplifies the process of sharing and reusing knowledge across different AI tools, making it easier for developers to build and maintain a cohesive and productive development environment. Repository: Asymptote-Labs/agent-beacon GitHub stars: 2295 GitHub description: The cross-harness, self-improving memory layer for AI agents. Join our community: https://discord.com/invite/zdNChS2fBu

### Overview: TorchLeet is a comprehensive collection of 65 machine learning and artificial intelligence interview problems designed specifically for developers. The repository includes Jupyter notebooks, an auto-grader, and an AI tutor to help you practice and improve your skills. ### Key Features: - **65 PyTorch Problems**: Covers a wide range of topics including neural networks, image classification, natural language processing, and more. - **Auto-Grader**: Provides immediate feedback on your solutions, helping you learn from mistakes. - **AI Tutor**: Offers a customizable learning path with hints, company-specific interview questions, and advanced explanations. - **Track Focus**: Divides problems into three tracks: Basics, LLM Learning Path, and Advanced, with overlapping questions for better practice. - **Quick Start**: Guides you through the basics of PyTorch and provides a simple way to start solving problems. ### Tech Stack: - **Jupyter Notebook**: The primary development environment for coding and experimenting with solutions. ### Why It Matters: - **Real-World Practice**: The problems are based on actual interview questions from top tech companies like Google, Meta, Anthropic, and more. - **Comprehensive Coverage**: Includes both beginner and advanced topics, ensuring a well-rounded understanding. - **Interactive Learning**: The AI tutor offers personalized learning paths and hints, making the practice more engaging and effective. ### GitHub Stars: - **2,579 Stars**: Demonstrates the popularity and usefulness of the repository among the developer community. ### Repository: - **Repository Name**: Exorust/TorchLeet

### Overview: `ort` is a Rust interface designed for fast hardware-accelerated inference and training of machine learning models in the ONNX format. Leveraging Microsoft's ONNX Runtime library, it supports a variety of hardware accelerators, making it ideal for deployment on devices or in the data center. ### Key Features: - **Hardware Acceleration**: Supports ONNX Runtime and other pure-Rust runtimes, ensuring efficient performance. - **Lightweight**: Runs efficiently on user devices, making it accessible on a wide range of platforms. - **Versatility**: Compatible with various hardware accelerators, including NVIDIA, AMD, and others. - **Performance**: Delivers fast inference times, making it suitable for real-time applications. - **Documentation**: Comprehensive documentation available, including guides, API references, and examples. ### Tech Stack: - Rust ### Why It Matters: `ort` simplifies the process of deploying machine learning models on-device or in the data center, making it an essential tool for developers working in the field of artificial intelligence and machine learning. ### GitHub Stars: 2,573 ### Repository: - **Repository**: `pykeio/ort` - **Primary Language**: Rust - **GitHub Stars**: 2,573 ### Documentation: - **Guide**: [https://ort.pyke.io/](https://ort.pyke.io/) - **API Reference**: [https://docs.rs/ort/2.0.0-rc.15/ort/](https://docs.rs/ort/2.0.0-