Projects

Overview: CareGPT is a medical language model that integrates multiple publicly available medical fine-tuned datasets and open-source medical large language models. It aims to accelerate the development of medical LLMs through comprehensive training, evaluation, and deployment. Key features: - Fine-tuning with ChatGPT - Deployment of ChatGPT-Next-Web - Gradio deployment - Training of LLaMA and LLaMA-2 - Support for LoRA and QLoRA - Integration with knowledge base - Open-source over 60 hospital department guide materials - Developed a tool for distilling medical data from GPT-4/ChatGPT models - Comprehensive collection of open-source medical LLMs, training data, deployment resources, and evaluation materials Tech stack: Python Why it matters: CareGPT is a significant contribution to the field of medical language models, aiming to accelerate the development of these models through comprehensive training, evaluation, and deployment. It leverages open-source resources to create a robust and accessible platform for medical LLMs, making it easier for researchers and practitioners to contribute to and benefit from this technology.

Overview: OML is a PyTorch-based framework designed for training and validating models that produce high-quality embeddings for metric learning and retrieval pipelines. Key features: - Supports PyTorch-based training and inference - Implements various metric learning algorithms - Provides a model zoo with pre-trained embeddings - Includes a documentation repository for easy access to the latest information Tech stack: Python Why it matters: OML offers a comprehensive solution for metric learning and retrieval tasks, making it a valuable tool for researchers and practitioners in the field.

### Overview: This repository is a curated list of papers and resources related to large language models on graphs, based on our survey paper "Large Language Models on Graphs: A Comprehensive Survey" (arXiv:2312.02783). It aims to provide a comprehensive review of scenarios and techniques related to large language models on graphs, focusing on their applications in various domains such as academic networks, e-commerce networks, and molecules with descriptions. ### Key Features: - **Pure Graphs**: - Datasets: - Examples include citation networks, social networks, and academic collaboration networks. - Direct Answering: - Techniques for answering questions directly using graph structures. - Heuristic Reasoning: - Methods for reasoning based on patterns and heuristics. - Algorithmic Reasoning: - Techniques for solving problems using algorithmic approaches. - **Text-Attributed Graphs**: - Datasets: - Examples include protein-protein interaction networks, chemical compounds, and citation networks. - LLM as Predictor (Node): - Methods for predicting node labels or attributes using LLMs. - LLM as Encoder: - Techniques for encoding graph structures into text representations. - LLM as Aligner (Node): - Methods for aligning graph structures with LLM predictions. - **Text-Paired Graphs (Molecules)**: - Datasets: - Examples include molecular graphs with descriptions. - LLM as Predictor (Graph

Overview: BS-RoFormer is an implementation of Band Split Roformer, a state-of-the-art attention network designed for music source separation. This model has significantly outperformed previous methods, particularly in the field of audio source separation. Key features: - Band Split Roformer: A multi-band attention network that splits the input audio into frequency bands for better separation. - Rotary Positional Encoding: An improvement over learned absolute positions, leading to better performance. - Support for Stereo Training: The model can be trained and evaluated on stereo audio inputs. - Multiple Stem Output: The model outputs multiple audio stems, allowing for more detailed separation. Tech stack: Python Why it matters: BS-RoFormer represents a significant advancement in the field of audio source separation, offering improved performance and flexibility compared to previous methods. Its ability to handle stereo inputs and produce multiple audio stems makes it highly valuable for various applications in music production and research. Repository: lucidrains/BS-RoFormer Primary language: Python GitHub stars: 994 GitHub description: Implementation of Band Split Roformer, SOTA Attention network for music source separation out of ByteDance AI Labs README excerpt: <img src="./bs-roformer.png" width="450px"></img>

### Overview: HOL Guard is an open-source antivirus solution designed to protect AI agents, tools, plugins, skills, and package installations from potential risks. It ensures secure runtime by reviewing actions before they execute, detecting and blocking risky behaviors like secret exposure, destructive operations, and prompt injection. ### Key Features: - **Secret Exposure Detection**: HOL Guard identifies and blocks exposure of sensitive information. - **Destructive Operations**: It prevents harmful actions that could compromise the system. - **Prompt Injection Prevention**: HOL Guard detects and prevents malicious prompts from being executed. - **Supply-Chain Risks**: It monitors and blocks supply-chain risks to prevent malicious packages from being installed. ### Tech Stack: - Python 3.10+ - Pipx for easy installation ### Why it Matters: HOL Guard is crucial for maintaining the integrity and security of AI agents and their associated tools, ensuring that they operate within safe and controlled environments. This tool helps prevent unauthorized actions that could lead to data breaches, system compromise, or other security incidents.

Overview: LLM-Blender is an innovative ensembling framework that leverages the strengths of multiple open-source large language models (LLMs) to achieve consistently superior performance. It addresses the issue of optimal LLMs for different examples by using a specialized pairwise comparison method and fusing the strengths of the top-ranked candidates. Key features: - LLM-Blender is an ensembling framework. - It cuts the weaknesses through pairwise ranking. - It integrates the strengths through generative fusion. Tech stack: Python Why it matters: LLM-Blender significantly outperforms state-of-the-art methods on various metrics across different benchmarks, including MixInstruct, establishing a substantial performance gap.

### Overview: The Perception package is a comprehensive toolkit designed for generating large-scale synthetic datasets in Unity, specifically for computer vision training and validation. It simplifies the process of creating and managing datasets, making it easier for developers to train models without the need for expensive real-world data. ### Key Features: - **Synthetic Data Generation**: Creates realistic synthetic environments for training and validation. - **Domain Randomization**: Adds variability to the synthetic data to simulate real-world conditions. - **Visualization Tools**: Provides tools to easily visualize and analyze generated datasets. - **Integration with Unity**: Seamlessly integrates with Unity's ecosystem, making it easy to incorporate into existing projects. - **Community Support**: Offers a mailing list for staying updated on the latest features and community events. ### Tech Stack: Unity C# ### Why It Matters: The Perception package is crucial for developers looking to accelerate the development of computer vision models without the financial burden of acquiring real-world data. It streamlines the process of creating and managing datasets, making it accessible to both beginners and experienced developers.

Overview: SurveyX is an academic survey automation system that leverages large language models to generate high-quality, domain-specific papers and surveys. It provides an offline processing capability, allowing users to request surveys based on uploaded references and keywords. Key features: - Generates comprehensive academic papers or surveys tailored to specific topics. - Supports real-time online search and access to a paper database. - Includes multimodal document parsing, including image understanding and illustrations. Tech stack: Python, LaTeX, Hugging Face, and OpenAI API. Why it matters: SurveyX simplifies the process of generating academic papers and surveys, making it accessible to researchers and students without requiring extensive knowledge of academic writing or technology.

### Overview: This repository contains a curated collection of eBooks on Data Science, Machine Learning, and AI, designed to help developers learn and grow in these fields. ### Key Features: - **Data Science and Machine Learning eBooks**: Comprehensive guides and tutorials on various Data Science and Machine Learning topics. - **AI Resources**: Articles and tutorials on Artificial Intelligence and related technologies. - **Learning Path**: A structured collection that can be used as a learning path or as a resource for specific projects. ### Tech Stack: Markdown, GitHub Pages, and possibly some basic HTML/CSS for formatting. ### Why it Matters: This repository serves as an invaluable resource for developers looking to enhance their skills in Data Science, Machine Learning, and AI. It provides a structured and curated way to learn from experts, making the learning process more effective and engaging.

Overview: KerasHub is a pretrained modeling library that simplifies the process of using popular model architectures with Keras 3, offering pre-trained checkpoints from Kaggle Models for various tasks like text, image, and audio data. Key features: - Pre-trained models for text, image, and audio data - Keras 3 compatible implementations - Pre-trained checkpoints from Kaggle Models - Supports JAX, TensorFlow, and PyTorch - Fine-tuning on individual accelerators and model/data parallel training - Easy installation via pip Tech stack: Python Why it matters: KerasHub streamlines the process of using pre-trained models for various tasks, making it easier and faster to get started with deep learning projects.

### Overview: soynlp 是一个基于纯 Python 的韩国语自然语言处理库，旨在通过学习数据来提取单词、分词和词性标注等功能。它提供了一系列的工具和功能，包括单词提取、分词器、词性标注器和向量器等，旨在帮助开发者进行文本分析和处理。 ### Key features: - **单词提取 (WordExtractor)**: 提取文本中出现的单词。 - **分词器 (Tokenizer)**: 将文本分割成单词。 - **词性标注器 (Part-of-Speech Tagger)**: 标注文本中的单词所属的词性。 - **向量器 (Vectorizer)**: 将文本转换为向量表示。 - **分隔符提取器 (Soyspacing)**: 提取文本中的分隔符。 - **关键词提取器 (Soykeyword)**: 提取文本中的关键词。 - **新闻名词提取器 (News Noun Extractor)**: 提取新闻文本中的名词。 - **分词器 (LTokenizer)**: 使用最小长度和最大长度的分词器。 - **分词器 (MaxScoreTokenizer)**: 使用最大分数的分词器。 - **分词器 (RegexTokenizer)**: 使用正则表达式的分词器。 ### Tech stack: Python 3.5+，支持numpy、psutil、scipy、scikit-learn等库。 ### Why it matters: soynlp 是一个强大的工具，可以帮助开发者快速地进行文本分析和处理，

Overview: The Multi-Agent Medical Assistant is an AI-powered chatbot designed for healthcare professionals, researchers, and patients. It leverages large language models, computer vision, retrieval augmented generation, real-time web search, and human-in-the-loop validation to provide medical assistance and research insights. Key features: - 🧑‍⚕️ Medical Diagnosis and Research Assistance - 📖 Document Processing and Retrieval - 🖼️ Medical Imaging Analysis - 🔍 RAG for Up-to-Date Medical Insights - 🚀 Real-time Web Search for Latest Medical Information Tech stack: Python Why it matters: This project showcases advanced AI techniques for medical diagnostics and research, making it valuable for healthcare professionals and researchers. Repository: souvikmajumder26/Multi-Agent-Medical-Assistant GitHub stars: 991 GitHub description: 🚕 GenAI powered multi-agentic medical diagnostics and healthcare research assistance chatbot. 🏥 Designed for healthcare professionals, researchers, and patients.

Overview: OpenTabs is a browser automation tool that allows developers to interact with web APIs using their authenticated browser sessions. It simplifies the process of testing and interacting with various services without the need for API keys or OAuth. Key features: - Supports multiple plugins and tools for various applications like Slack, Discord, GitHub, Jira, Notion, Figma, AWS, and Stripe. - Built-in browser tools for screenshots, clicks, typing, and network capture. - Built-in plugin development capabilities to scaffold plugins, publish them to npm, and install them for use. Tech stack: TypeScript, Node.js, and the Chrome extension. Why it matters: OpenTabs simplifies the process of interacting with web APIs by abstracting away the complexity of managing API keys and OAuth, making it easier for developers to test and automate their interactions without the need for additional setup.

### Overview: Claude Skills are a comprehensive collection of 107 stack-agnostic skills designed to cover the full website lifecycle. These skills are opinionated and provide a consistent framework for building, launching, running, and growing a brand and website. They include an Ahrefs MCP-powered SEO audit suite and can be used on Next.js, WordPress, Shopify, Webflow, plain HTML, or any other platform. ### Key Features: - **Brand, Design, Content, SEO, Dev, Ops, Growth, and Research Skills**: Cover the entire website development and marketing lifecycle. - **SEO Audit Suite**: Utilizes Ahrefs MCP for comprehensive SEO audits. - **Cross-Platform Compatibility**: Works seamlessly across Claude.ai, Claude Code, and the Anthropic API. - **Skill Catalog**: A curated list of 107 skills, each with a brief description and instructions. ### Tech Stack: - **JavaScript**: The primary language used to develop and implement these skills. ### Why It Matters: These skills provide a structured and consistent approach to website development and marketing, ensuring that Claude can handle a wide range of tasks across different platforms and technologies. They offer a valuable resource for developers looking to streamline their workflow and improve their skills in various areas of website management and marketing. ### Repository: The repository for this project is hosted on GitHub, where you can find the project's primary language (JavaScript) and the number of GitHub stars (986). The README provides detailed instructions on how to install the skills in Claude Code and offers a table of contents for easy navigation. ###

### Overview: This repository contains a monthly series of the Top 10 Open Source Projects in Machine Learning, selected from new or major releases in the field. The series is curated by Mybridge AI, which ranks articles based on shares, minutes read, and its own machine learning algorithm. Each month, 100 to 300 projects are compared, and only the finest 10 are selected. ### Key features: - **Machine Learning Projects**: Monthly curated list of top 10 open source projects in Machine Learning. - **Machine Learning Open Source Monthly**: Monthly series focusing on Machine Learning open source projects. - **GitHub Stars**: 992 stars, indicating the project's popularity among developers. - **GitHub Description**: Monthly Series - Machine Learning Top 10 Open Source Projects. - **Readme Excerpt**: Click "Watch/Star" to get an email notification once a month for Top 10 Machine Learning Open Source. ### Tech Stack: The project uses Markdown and GitHub Pages to organize and display the monthly series. ### Why it Matters: This repository serves as a valuable resource for developers interested in Machine Learning, providing insights into the latest and finest open source projects in the field. It helps users stay updated with the latest trends and best practices in Machine Learning development.

### Overview: Deep Image Matting is a project aimed at reproducing and improving the Deep Image Matting algorithm. The repository includes all necessary dependencies and datasets required for training and testing the model. It utilizes TensorFlow and Keras for implementation, and is designed to be used in various image processing tasks. ### Key Features: - **Dependencies**: NumPy, TensorFlow 1.9.0, Keras 2.1.6, and OpenCV. - **Datasets**: Adobe Deep Image Matting Dataset, MSCOCO, and PASCAL VOC. - **ImageNet Pretrained Models**: VGG16 weights. - **Usage**: Data pre-processing, training, visualization during training, and a demo script for running the model. ### Tech Stack: - Python ### Why It Matters: Deep Image Matting is a significant algorithm for image editing and manipulation, particularly in the field of image matting. It allows for the separation of foreground and background in images, which can be useful in various applications such as image editing, content-aware compositing, and more. ### Repository: - **Primary Language**: Python - **GitHub Stars**: 991 - **GitHub Description**: Deep Image Matting ### Usage: - **Data Pre-processing**: Extract training images using `pre_process.py`. - **Train**: Run `train.py` to train the model. - **Visualization**: Run `tensorboard --logdir path_to_current_dir/logs` to visualize training progress. - **Demo**: Download pre-trained model and run `demo.py` to see the

Overview: This repository contains a PyTorch implementation of the Center Loss algorithm, which is used in deep face recognition and other applications. It is based on the paper "A Discriminative Feature Learning Approach for Deep Face Recognition" by Wen et al., published at ECCV 2016. Key features: - PyTorch implementation of the Center Loss algorithm - Used in deep face recognition and related applications - Based on the paper "A Discriminative Feature Learning Approach for Deep Face Recognition" by Wen et al., published at ECCV 2016 Tech stack: Python Why it matters: The Center Loss algorithm is crucial in deep face recognition and related applications, providing a way to learn discriminative features that are invariant to the location and scale of the input images. This makes it particularly useful in scenarios where the input images can vary significantly. GitHub stars: 990 Repository: KaiyangZhou/pytorch-center-loss

Overview: AI-Compass 是一个全面覆盖人工智能技术栈的综合性开源项目，旨在为开发者提供系统性地了解 AI 的核心概念、主流技术、前沿趋势，并通过实践掌握从理论到落地的全过程的资源和指南。 Key features: - 系统化组织的 AI 技术文章和博客 - 实战代码和项目展示 - 从基础理论到前沿应用的完整知识体系 - 丰富的工程实践资源 - 企业级开源资源 Tech stack: Python Why it matters: AI-Compass 通过整合从基础理论到前沿应用的完整知识体系，帮助开发者系统性地了解 AI 的核心概念、主流技术、前沿趋势，并通过实践掌握从理论到落地的全过程。项目还包含了丰富的工程实践资源，从模型训练、推理部署到产品化落地的全流程技术指南，帮助开发者构建从概念验证到商业化应用的完整技术能力。 GitHub stars: 983

### Overview: MLX Studio is a native macOS app that simplifies the process of running large language models, vision models, and image generation on Apple Silicon. It wraps the vMLX inference server, providing a seamless experience without requiring any setup or configuration. ### Key Features: - **Chat Interface**: Multi-turn conversations with streaming, markdown rendering, and code highlighting. - **Vision**: Send images to multimodal models (Qwen-VL, LLaVA, Pixtral, and more). - **Reasoning**: Native support for thinking/reasoning models (DeepSeek-R1, QwQ, GLM-Z1). - **Tool Calling**: Function calling with structured output for agent workflows. - **Image Generation**: Text-to-image with MLX-powered diffusion models. - **Text-to-Speech**: TTS via Kokoro with multiple voices. - **Speech-to-Text**: Whisper-based transcription. - **OpenAI-Compatible API**: HTTP server on `localhost` for any compatible client. - **Menu Bar Mode**: Runs in the tray for quick access. - **Developer Tools**: Model inspection, conversion, quantization, and diagnostics. - **Session Management**: Save, restore, and organize conversations with SQLite persistence. - **JANG Quantization**: First-class support for [JANG mixed-precision quantization](https://github.com/jjang-ai/vmlx) formats. ### Tech Stack: Apple's MLX framework for GPU-accelerated inference. ### Why It Matters: MLX Studio simplifies the process of running large language models and vision models on

Overview: CodeProject.AI Server is a self-contained, open-source AI microserver designed to be integrated into software applications, providing developers with the power of AI without requiring off-device or network data transfer. Key features: - Real-time object detection - Natural language processing - Image classification - Speech recognition Tech stack: C# Why it matters: This project aims to simplify the integration of AI into software applications, making it accessible and easy to use for developers, while also providing a platform for community contributions and improvements.