Projects

Overview: This repository contains the code and scripts for training and evaluating a state-of-the-art conversational AI model using transfer learning techniques. The model is based on OpenAI's GPT and GPT-2 transformers and is designed to be easily installed and used for both single and distributed training on a cloud instance. Key features: - Training and testing scripts for a conversational AI model - Pre-trained and fine-tuned model available on S3 - Installation instructions for both local and Docker-based environments - Training script supports single GPU and distributed multi-GPU settings - Arguments for adjusting training parameters such as dataset, model, and learning rate Tech stack: Python Why it matters: This project showcases the potential of transfer learning in improving conversational AI models, demonstrating how pre-trained models can be fine-tuned for better performance on specific tasks.

### Overview: CLAM is a high-throughput and interpretable method for data efficient whole-slide image (WSI) classification using slide-level labels without any ROI extraction or patch-level annotations. It is capable of handling multi-class subtyping problems and is tested on three different WSI datasets. ### Key features: - Data efficient and weakly supervised whole-slide image classification. - Slide-level labels for high-throughput classification. - Multi-class subtyping capability. - Tested on three different WSI datasets. - Capable of handling independent test cohorts of WSI resections and biopsies. - Smartphone microscopy images (photomicrographs) as well as WSI datasets. ### Tech stack: Python ### Why it matters: CLAM is a significant advancement in computational pathology, offering a high-throughput and interpretable method for whole-slide image classification. It addresses the challenges of data efficiency and weak supervision, making it suitable for various applications in the field of computational pathology.

### Overview: 整理汇总了2026年、2025年、2024年、2021年和2020年CVPR底层视觉(Low-Level Vision)相关的一些论文和代码，包括超分辨率、图像去雨、图像去雾、去模糊、去噪、图像恢复、图像增强、图像去摩尔纹、图像修复、图像质量评价、插帧、图像/视频压缩等任务。 ### Key features: - 超分辨率 - 图像去雨 - 图像去雾 - 去模糊 - 去噪 - 图像恢复 - 图像增强 - 图像去摩尔纹 - 图像修复 - 图像质量评价 - 插帧 - 图像/视频压缩 ### Tech stack: Unknown ### Why it matters: 整理汇总了2026年、2025年、2024年、2021年和2020年CVPR底层视觉(Low-Level Vision)相关的一些论文和代码，包括超分辨率、图像去雨、图像去雾、去模糊、去噪、图像恢复、图像增强、图像去摩尔纹、图像修复、图像质量评价、插帧、图像/视频压缩等任务。

Overview: This project predicts NBA game winners and totals using machine learning models trained on team stats and sportsbook odds. It includes features for moneyline and totals predictions, expected value calculation, and optional Kelly Criterion sizing. Key features: - Moneyline and totals predictions (XGBoost and Neural Net models) - Expected value calculation and optional Kelly Criterion sizing - Odds ingest from supported sportsbooks or manual input - Data processing pipeline and model training scripts - Flask web app for browsing outputs Tech stack: Python Why it matters: This project leverages machine learning to enhance sports betting decisions, offering valuable insights into game outcomes and potential payouts. Repository: kyleskom/NBA-Machine-Learning-Sports-Betting Primary language: Python GitHub stars: 1755 GitHub description: NBA sports betting using machine learning

### Overview: A collection of resources and tools for AIOps, including white papers, courses, industry practices, articles, algorithms, datasets, and useful WeChat official accounts. The repository aims to provide a comprehensive guide for developers interested in AIOps. ### Key features: - White Paper: Comprehensive guidance on implementing AIOps. - Course and Slides: Courses and slides from Tsinghua University on AIOps. - Industry Practice: Real-world examples from major tech companies. - Article: Articles on AIOps, including mapping approaches and KPI anomaly detection. - Tools and Algorithms: Various tools and algorithms for monitoring and visualizing microservices architecture, such as python-fp-growth, Anomaly Detection with Twitter in R, and Curve. - Dataset: Real-world datasets for anomaly detection. - Useful WeChat Official Accounts: Official accounts providing insights into AIOps. ### Tech stack: Python, R, JavaScript, and various open-source tools. ### Why it matters: This repository serves as a valuable resource for developers looking to enhance their AIOps skills, providing both theoretical knowledge and practical tools to implement and improve AIOps practices in their projects.

Overview: Maximal Update Parametrization (μP) and Hyperparameter Transfer (μTransfer) is a technique for tuning hyperparameters of extremely large neural networks, specifically large pretrained transformers. This method ensures stable hyperparameter stability across neural network sizes, making it useful for tuning large models. Key features: - Maximal Update Parametrization (μP) - Hyperparameter Transfer (μTransfer) - Seamless integration with PyTorch models Tech stack: Jupyter Notebook Why it matters: μP reduces the fragility and uncertainty when transitioning from exploration to scaling up, addressing issues not often discussed in the deep learning literature. GitHub stars: 1756 Repository: microsoft/mup

### Overview: Agent Development Kit (ADK) for Java is an open-source, code-first toolkit designed for developers aiming for fine-grained control and flexibility when building advanced AI agents. It seamlessly integrates with Google Cloud services, enabling agents to be tightly integrated with the ecosystem. ### Key Features: - **Rich Tool Ecosystem**: Utilize pre-built tools, custom functions, OpenAPI specs, or integrate existing tools for diverse capabilities. - **Code-First Development**: Define agent logic, tools, and orchestration directly in Java for ultimate flexibility, testability, and versioning. - **Modular Multi-Agent Systems**: Design scalable applications by composing multiple specialized agents into flexible hierarchies. ### Tech Stack: Java ### Why it Matters: ADK provides developers with the ability to build, evaluate, and deploy sophisticated AI agents with tight integration and control, making it ideal for applications that require seamless integration with Google Cloud services.

Overview: FARM is a fast and easy transfer learning framework designed for natural language processing (NLP). It simplifies the process of integrating language models into various industries, focusing on question answering. Key features: - Fast and efficient transfer learning - Industry-specific applications for question answering - Easy integration with existing NLP models Tech stack: Python Why it matters: FARM simplifies the complex process of integrating language models into industries, making it accessible and efficient for developers and researchers alike. Repository: deepset-ai/FARM Primary language: Python GitHub stars: 1754

### Overview: ToolUniverse is a powerful platform that democratizes AI research by providing a user-friendly interface for scientists to access and utilize state-of-the-art AI tools and techniques. ### Key Features: - **AI Agent Setup**: A simple agent setup guide is provided, making it easy for AI agents to integrate ToolUniverse into their workflows. - **Manual Setup**: A detailed guide for setting up ToolUniverse manually, including adding skills and configuring the MCP. - **Python SDK**: A Python SDK is available for developers to integrate ToolUniverse into their projects. - **CLI Interface**: A CLI tool, `tu`, is provided for quick access to the most frequently used features. - **Documentation**: Comprehensive documentation is available on the website, including installation guides and API references. ### Tech Stack: Python ### Why it Matters: ToolUniverse empowers researchers and scientists by providing a unified interface for accessing cutting-edge AI tools and techniques, making it easier to collaborate and innovate without the need for extensive technical knowledge. ### Repository: mims-harvard/ToolUniverse ### GitHub Stars: 1754 ### GitHub Description: Democratizing AI scientists with ToolUniverse

### Overview: Auto-Deep-Research is an open-source and cost-efficient alternative to Deep Research, designed to be a fully-automated and cost-effective personal AI assistant. It leverages the AutoAgent framework for seamless integration with various Large Language Models (LLMs) and supports both function-calling and non-function-calling interactions. ### Key Features: - **High Performance**: Delivers good performance on the GAIA Benchmark. - **Universal LLM Support**: Seamlessly integrates with a wide range of LLMs, including OpenAI, Anthropic, Deepseek, vLLM, Grok, and Huggingface. - **Flexible Interaction**: Supports both function-calling and non-function-calling interactions with LLMs. - **Cost-Efficient**: Offers an open-source alternative to Deep Research's $200/month subscription, with pay-as-you-go LLM API keys. - **File Support**: Handles file uploads for enhanced data interaction. - **One-Click Launch**: Instantly get started with a simple `auto deep-research` command, requiring zero configuration. ### Tech Stack: Auto-Deep-Research is built on Python. ### Why It Matters: Auto-Deep-Research provides a cost-effective and highly performant solution for personal AI assistants, making it accessible to a broader audience without the high costs associated with commercial alternatives.

Overview: The Alan AI SDK for Flutter enables developers to create intelligent apps by embedding an AI agent into their Flutter applications, providing context-aware code generation and real-time feature creation. Key features: - Embedding an AI agent into Flutter applications - Real-time generation of business logic and UI - Integration with the Alan AI Platform and Studio - Support for multiple platforms including Web, Android, iOS, Ionic, Apache Cordova, React Native, and PowerApps Tech stack: Flutter, Dart, Alan AI Platform, Alan AI Studio, Three-Layer AI (3LAI) architecture Why it matters: The Alan AI SDK for Flutter simplifies the integration of AI-driven interfaces into existing apps, enabling companies to create adaptive software that responds and evolves automatically, enhancing user experience and business outcomes.

### Overview: Bespoke Curator is a powerful tool for synthetic data curation and post-training data extraction, designed to streamline the process of fine-tuning models and generating high-quality reasoning datasets. ### Key Features: - **Synthetic Data Curation**: Automatically curates and prepares synthetic data for post-training fine-tuning. - **Scalable Data Extraction**: Extracts structured data efficiently from various sources. - **Fine-Tuning with Tinker**: Integrates with the Tinker SDK for seamless fine-tuning of models. - **Data Integration with Fireworks AI**: Supports fine-tuning using Fireworks AI for supervised learning. - **OpenThoughts-Agents**: Curates data for the OpenThoughts-Agents project, which won a $5000 USD prize. - **OpenThoughts2-1M Dataset**: Created using Bespoke Curator, this dataset was used to train OpenThinker2-32B, outperforming DeepSeek-R1-32B. - **Gemini Batch Support**: Simplifies Gemini batch processing, making it easier to handle large-scale data. - **Code Executor**: Runs code generated by Curator using multiple backends, including local, Ray, Docker, and e2b. ### Tech Stack: Python ### Why It Matters: Bespoke Curator is essential for researchers and developers looking to streamline their data processing and fine-tuning workflows, ensuring efficient and effective use of synthetic data and model training.

Overview: CC Safety Net is a pre-execution guard designed to protect AI coding agents from destructive Git and file system commands, as well as attempts to access sensitive files. It ensures safe tool calls by parsing command inputs and blocking dangerous operations. Key features: - Blocks destructive commands and access to secrets - Parses command inputs to prevent command obfuscation - Supports various coding agent CLIs on Windows, macOS, and Linux - Does not contain processes, set filesystem permissions, or watch network egress Tech stack: TypeScript Why it matters: CC Safety Net enhances security by preventing harmful commands from running, ensuring the integrity and safety of the development environment. Repository: kenryu42/cc-safety-net GitHub stars: 1749

Overview: llm-ui is a React library designed specifically for integrating Large Language Models (LLMs) into web applications. Key features: - Removes broken markdown syntax - Allows customization of LLM output - Smooths out pauses in LLM streaming - Renders output at native frame rate - Supports code blocks for various languages - Headless, enabling styles to be brought in from the developer Tech stack: TypeScript Why it matters: llm-ui simplifies the integration of LLMs into React applications, enhancing user experience with native rendering and customization options. Repository: richardgill/llm-ui Primary language: TypeScript GitHub stars: 1749 GitHub description: The React library for LLMs README excerpt: The React library for LLMs. [Documentation](http://llm-ui.com/docs) [Sponsor](https://getstream.io/?utm_source=github.com&utm_medium=github&utm_campaign=llm-ui)

Overview: Risuai is a user-friendly software for LLM roleplaying, offering multiple API supports, emotion images, group chats, plugins, regex functions, powerful translators, lorebook, themes, customizable UI, TTS, additional assets, long-term memory, and more. Key features: - Multiple API Supports - Emotion Images - Group Chats - Plugins - Regex Script - Powerful Translators - Lorebook - Themes - Customizable, Friendly UI - TTS - Additional Assets - Long-term Memory Tech stack: TypeScript Why it matters: Risuai provides a comprehensive and user-friendly platform for roleplaying with advanced features, making it accessible and powerful for LLM enthusiasts and roleplayers alike. Repository: kwaroran/Risuai Primary language: TypeScript GitHub stars: 1747 GitHub description: Make your own story. User-friendly software for LLM roleplaying

### Overview: 书童 Pagekin 是一款开源免费的 AI 长篇小说写作桌面软件，专为长篇小说和网文连载设计。它支持从设定、蓝图到正文、审稿、定稿的全流程操作，并提供拆解仿写、续写、改写功能，本地优先处理，不带模型。软件界面简洁，支持多种语言和主题色，可自定义写作技能。 ### Key features: - **本地优先处理**：作品存放在用户自己的电脑上，不依赖任何模型。 - **从设定到定稿**：从设定、架构、蓝图到正文、审稿、定稿的全流程支持。 - **拆解仿写、续写、改写**：支持导入原稿续写、段落续写、改写功能。 - **角色管理**：角色档案和角色列表编辑，角色状态和关系图谱。 - **章节蓝图**：逐章列出规划，支持批量创作和暂停。 - **正文编辑**：支持键盘输入和全角空格缩进，逐章输出实时显示。 - **运行侧页**：实时输出生成任务，支持取消和暂停批量创作。 - **作品资料与版本**：剧情与伏笔进度、知识库、版本历史、导出备份。 - **写作 Skill**：支持导入自定义写作技能，如去 AI 味、文风等。 - **深色主题与中英文界面**：支持浅色和深色主题，支持中文和英文

Overview: This repository contains a collection of projects, tools, and research papers related to deepfakes, focusing on ethical use and defense against deepfake manipulation. Key features: - Faceswap - Faceswap2 - Faceit - DeepFaceLab - DeepfakeCapsuleGAN - Large resolution facemasked - Disrupting Deepfakes Tech stack: Python, TensorFlow, PyTorch, OpenCV, and other relevant deep learning frameworks. Why it matters: This repository serves as a resource for developers and researchers interested in the field of deepfakes, providing tools and knowledge to enhance and promote ethical research and development.

### Overview: BrushNet is a state-of-the-art image inpainting model that leverages diffusion-based text-guided techniques. It is designed to be easily integrated into existing diffusion models, making it highly versatile and efficient for various image inpainting tasks. ### Key Features: - **Plug-and-Play Integration**: BrushNet can be seamlessly integrated into any pre-trained diffusion model, making it highly versatile. - **Text-Guided Inpainting**: It uses text prompts to guide the inpainting process, ensuring high-quality results. - **Efficient Training**: The model is trained on a reduced dataset, making it faster and more efficient to train compared to full pre-trained models. ### Tech Stack: - Python 3.9 - PyTorch 1.12.1 ### Why It Matters: BrushNet offers a powerful solution for image inpainting tasks, providing high-quality results with minimal effort. Its modular design and efficient training process make it an excellent choice for researchers and practitioners alike. ### Repository: Repository: [TencentARC/BrushNet](https://github.com/TencentARC/BrushNet) ### GitHub Stars: GitHub Stars: 1745 ### GitHub Description: This repository contains the implementation of the ECCV2024 paper "BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion".

### Overview: MobileNetV2-YoloV3-Nano is a lightweight model designed for mobile devices, offering a 0.5BFlops performance with a 3MB weight size. It achieves 73.26% VOC mAP and 37.44% COCO mAP, making it highly efficient for real-time applications. ### Key Features: - **Lightweight Model**: 0.5BFlops, 3MB weight size. - **High Performance**: Inference time of 10.16 ms for MobileNetV2-YOLOv3-Nano. - **Mobile Inference Frameworks**: Supports NCNN and MNN, with benchmarks provided for both. - **Inference Time and FLOPs**: Inference times and FLOPs for different models are provided. - **MobileNetV2-YOLOv3-Lite and Nano**: Offers improved performance with lower inference times and FLOPs. ### Tech Stack: - **C** Programming Language. ### Why It Matters: This model is optimized for mobile devices, making it highly efficient and suitable for real-time applications such as face detection and human body posture estimation.

### Overview: Agentic Retrieval-Augmented Generation (Agentic RAG) explores advanced Retrieval-Augmented Generation systems enhanced with AI LLM agents. This repository provides insights into foundational principles, a detailed taxonomy of Agentic RAG systems, comparative analysis of traditional RAG, Agentic RAG, and Agentic Document Workflows (ADW), real-world applications, and challenges in scaling, ethical AI, multimodal integration, and human-agent collaboration. ### Key Features: - **Taxonomy of Agentic RAG Systems**: A comprehensive taxonomy showcasing frameworks like single-agent, multi-agent, hierarchical, corrective, adaptive, and graph-based RAG. - **Agentic Workflow Patterns**: Patterns for reflection, planning, tool use, and multi-agent collaboration. - **Comparative Analysis**: Analysis of traditional RAG, Agentic RAG, and Agentic Document Workflows (ADW) to highlight their strengths, weaknesses, and best-fit scenarios. - **Real-World Applications**: Applications across industries like healthcare, education, finance, and legal analysis. - **Challenges and Future Directions**: Challenges in scaling, ethical AI, multimodal integration, and human-agent collaboration. ### Tech Stack: Unknown ### Why it Matters: This repository serves as a comprehensive resource for researchers and practitioners to explore, implement, and advance the capabilities of Agentic RAG systems. It provides insights into foundational principles, a detailed taxonomy of Agentic RAG systems, comparative analysis of traditional RAG, Agentic RAG, and Agentic Document Workflows (ADW), real-world applications, and challenges in