Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

Python 356 stars

digital-go-jp/genai-ai-api

源内AIアプリ

Overview: 源内（GenAI）は、デジタル庁が開発・運用する生成 AI 利活用基盤です。行政職員が業務特化の生成 AI アプリケーションを、迅速かつ安全かつ簡単に利用できる環境を提供します。 Key features: - 行政実務用 AI アプリの公開 - 生成 AI を活用したマイクロサービス - 2種類のシステム構成：源内 Web と行政実務用 AI アプリ Tech stack: Python Why it matters: このリポジトリでは、デジタル庁が実際に展開している行政実務用 AI アプリの一部を公開しています。このリソースは、生成 AI を活用した行政業務の自動化と効率化を促進し、行政職員の業務負荷を軽減するための重要なツールです。 Repository: digital-go-jp/genai-ai-api GitHub stars: 356 GitHub description: 源内AIアプリ

Python 356 stars

stevenlsw/physgen

PhysGen: Rigid-Body Physics-Grounded Image-to-Video Generation (ECCV 2024)

### Overview: PhysGen is a novel image-to-video generation pipeline that integrates physical simulation and generative video diffusion prior. This project presents a method for generating videos from images using rigid-body physics, showcasing its effectiveness in ECCV 2024. ### Key Features: - **Image-to-Video Generation**: Converts static images into dynamic videos. - **Physics-Grounded**: Uses physical simulation to generate realistic videos. - **Training-Free**: The pipeline is training-free, eliminating the need for extensive training data. ### Tech Stack: - Python - PyTorch ### Why It Matters: PhysGen offers a unique approach to image-to-video generation by leveraging physical simulation, providing high-quality results without the need for extensive training data. This makes it particularly useful for applications requiring real-time performance and high fidelity. ### Repository: Repository: stevenlsw/physgen GitHub Stars: 356 GitHub Description: PhysGen: Rigid-Body Physics-Grounded Image-to-Video Generation (ECCV 2024) ### Installation: - Clone the repository: ```Shell git clone --recurse-submodules https://github.com/stevenlsw/physgen.git cd physgen ``` - Install requirements: ```Shell conda create -n physgen python=3.9 conda activate physgen pip install -r requirements.txt ``` ### Colab Notebook: - Run the Colab notebook for quick start: ```Shell export PYTHONPATH=$(pwd)

Unknown 356 stars

tonywu71/colpali-cookbooks

Recipes for learning, fine-tuning, and adapting ColPali to your multimodal RAG use cases. 👨🏻‍🍳

Overview: This repository contains notebooks for learning about the ColPali model, fine-tuning it for specific use cases, and creating similarity maps to interpret its predictions. Key features: - ColPali: A model designed for document retrieval using visual features. - ColQwen2: A model for inference and interpretability using the 🤗 transformers-native implementation. - ColPali: A model for fine-tuning and similarity map generation. Tech stack: 🤗 Transformers, 🤗 Hugging Face. Why it matters: This repository provides a comprehensive set of tools for developers to learn about and utilize the ColPali model, making it easier to integrate and optimize for various multimodal RAG use cases.

Python 355 stars

2dogsandanerd/Knowledge-Base-Self-Hosting-Kit

A Docker-powered RAG system that understands the difference between code and prose. Ingest your codebase and documentation, then query them with full privacy and zero configuration.

### Overview: This project is a Docker-powered RAG (Retrieval-Augmented Generation) system designed for developers. It allows users to ingest their codebase and documentation, then query them with full privacy and zero configuration. The system is built on Python and Docker, making it easy to set up and use. ### Key Features: - **Ingestion Modes:** Supports single uploads, folder scans, and hybrid chunking. - **Agent-Ready Search:** Enables semantic queries, citation viewing, and streaming logs. - **MCP Integration:** Connects with OpenClaw or other MCP-aware agents via the included MCP server. - **Agent Configuration Tab:** Configures MCP connection details without modifying `.env` or the CLI. - **Local LLM Friendly:** Supports both default Ollama setup and OpenAI-compatible servers. ### Tech Stack: Python ### Why it Matters: This system provides a robust and easy-to-use RAG solution for developers, allowing them to leverage their existing code and documentation without the need for complex configuration. It ensures privacy and security while providing powerful querying capabilities.

Python 355 stars

HenestrosaDev/audiotext

A desktop app to transcribe audio and video from files, URLs, YouTube videos, or your mic, then edit, translate, summarize, subtitle, and chat with your transcripts.

### Overview: Audiotext is a desktop application designed for transcribing audio and video files, YouTube videos, and microphone recordings into various languages. It supports 100 languages and can handle multiple formats including MP3, MP4, and WAV files, as well as YouTube videos and media file links. The app allows for editing, translating, summarizing, subtitling, and chatting with transcripts. ### Key features: - Transcription of audio and video files - Translation of transcriptions - Summarization of transcriptions - Subtitling of transcriptions - Chatting with transcripts ### Tech stack: - Python ### Why it matters: Audiotext is a powerful tool for language learners, content creators, and anyone needing to quickly transcribe audio and video content into multiple languages. Its versatility and ease of use make it a valuable addition to any professional or personal toolkit. ### Repository: HenestrosaDev/audiotext - GitHub stars: 354 - GitHub description: A desktop app to transcribe audio and video from files, URLs, YouTube videos, or your mic, then edit, translate, summarize, subtitle, and chat with your transcripts. ### Languages and Formats: - Supports 100 languages - Handles multiple formats including MP3, MP4, WAV, and YouTube videos ### Project Structure: - The project is structured to allow for easy installation and modification. ### Built With: - Python ### Getting Started: - **Installation**: The app can be installed using pip. - **Setting Up the Project

Jupyter Notebook 355 stars

MelihGulum/Comprehensive-Data-Science-AI-Project-Portfolio

A curated collection of AI, data engineering, and DevOps projects featuring real-world applications, advanced techniques, and tutorials—ideal for learners and practitioners exploring data science and machine learning.

### Overview: This repository is a curated collection of projects across various domains including Machine Learning, Deep Learning, Data Engineering, Data Analysis, AI Systems, and Cloud/MLOps. Each project folder includes detailed READMEs with setup instructions, methodologies, and results. It features end-to-end workflows, real-world datasets, deployment-ready pipelines, and reproducible best practices. ### Key features: - **Projects**: A selection of impactful and advanced projects. - **Machine Learning Projects**: Focus on classification and regression tasks using tools like CNN, LSTM, and Optuna. - **Deep Learning Projects**: Cover a range of tasks including image classification with MobileNetV2 and real-time face mask detection. - **Data Engineering Projects**: Include ETL pipelines for NBA player statistics and data scraping for web applications. - **AI Systems Projects**: Showcase systems like a university information retrieval system using RAG and LLM. - **Cloud/MLOps Projects**: Provide foundational knowledge on Terraform and AWS IaC. ### Tech stack: Python, Scikit-Learn, TensorFlow, SQL, Kafka, Docker, Terraform, Flask, AWS, GCP. ### Why it matters: This repository is ideal for learners and practitioners exploring data science and machine learning, offering real-world applications, advanced techniques, and practical tutorials.

Python 355 stars

MikeWangWZHL/Solo-Performance-Prompting

Repo for paper "Unleashing Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration"

Overview: This project explores the concept of cognitive synergy in large language models through a task-solving agent that collaborates with multiple personas. Key features: - Multi-persona self-collaboration - Emergent cognitive synergy - Task-solving capabilities Tech stack: Python Why it matters: This research demonstrates how different personas can work together to enhance the performance and creativity of large language models, showcasing the potential for improved collaboration and synergy in AI systems. Repository: MikeWangWZHL/Solo-Performance-Prompting Primary language: Python GitHub stars: 356 GitHub description: Repo for paper "Unleashing Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration"

Python 355 stars

PolyX-Research/Robust-U1

🚀🚀🚀 [ICML 2026] Official Implementation of Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding?

Overview: Robust-U1 is an MLLM model designed to recover corrupted visual content and reason over it, achieving robust visual understanding under real-world image degradations. Key features: - Robust-U1: Unified model for visual self-recovery and multimodal reasoning. - Robust-U1-SFT: Stage-I supervised fine-tuned checkpoint. - Robust-U1-RL: Stage-II reinforcement-learning checkpoint. Tech stack: Python. Why it matters: This model addresses the limitations of existing approaches in robust visual understanding, enabling faithful visual understanding under real-world image degradations. Repository: PolyX-Research/Robust-U1 Primary language: Python GitHub stars: 355 GitHub description: 🚀🚀🚀 [ICML 2026] Official Implementation of Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding?

Python 355 stars

lucidrains/bit-diffusion

Implementation of Bit Diffusion, Hinton's group's attempt at discrete denoising diffusion, in Pytorch

### Overview: Bit Diffusion is an implementation of Hinton's group's attempt at discrete denoising diffusion, focusing on the use of Pytorch. The repository aims to benefit the research community by providing a clean and accessible framework for further development. ### Key Features: - **Implementation of Bit Diffusion**: A discrete diffusion model inspired by Hinton's group's work. - **Pytorch Framework**: Utilizes Pytorch for efficient and streamlined implementation. - **Image Generation**: Generates discrete images from continuous data using the diffusion process. ### Tech Stack: Python ### Why it Matters: The clean and accessible repository will likely provide researchers with a solid foundation and benefits for those building upon this research.

Python 354 stars

JIA-Lab-research/VisionReasoner

[ICLR 2026] VisionReasoner: Unified Reasoning-Integrated Visual Perception via Reinforcement Learning

### Overview: VisionReasoner is a unified framework for visual perception tasks, demonstrating strong multi-task capability and superior performance across diverse visual perception tasks within a single unified model. This project builds upon the advancements in Vision Reasoning and aims to advance the frontier of Visual Language Models (VLMs) research. ### Key Features: 1. **Unified Framework**: VisionReasoner is a single VLM that can solve various vision tasks, showcasing its ability to handle diverse visual perception tasks within a shared model. 2. **Representative Tasks**: The model evaluates models' unified visual ability across tasks such as detection (e.g., COCO, RefCOCOg), segmentation (e.g., ReasonSeg), counting (e.g., CountBench), and VQA (e.g., DocVQA). 3. **Performance Superiority**: Experimental results show that VisionReasoner outperforms baseline models across ten diverse visual perception tasks, achieving significant improvements. ### Tech Stack: - **Python**: The primary programming language used for developing and training the model. ### Why It Matters: This project aims to advance the research and application of VLMs by demonstrating the potential of a single VLM to solve a wide range of vision tasks. It expands the capabilities of VLMs and opens up new possibilities in fields such as computer vision and natural language processing. ### Repository: Repository: JIA-Lab-research/VisionReasoner GitHub Stars: 354 GitHub Description: [ICLR 2026] VisionReasoner: Unified Reasoning-Integrated Visual Perception via Rein

HTML 354 stars

ddzipp/AutoAudit

AutoAudit—— the LLM for Cyber Security 网络安全大语言模型

Overview: AutoAudit is a cybersecurity LLM model designed to enhance threat detection, analysis, and response through advanced language processing capabilities. Key features: - AutoAudit-7B: A demo version trained on Alpaca-Lora, suitable for network security with satisfactory responses in English. - AutoAudit-8B-Instruct: Fine-tuned on Llama3-8B-instruct, significantly improved in cybersecurity-related query responses and foundational capabilities. - AutoAudit-Qwen: Still in the exploration and planning stage due to limited Chinese cybersecurity corpus. - More LLMs are coming soon. Tech stack: HTML Why it matters: LLMs offer a transformative approach to cybersecurity by providing advanced language processing capabilities, enhancing threat detection, analysis, and response, and improving the adaptability and resilience of security systems. GitHub stars: 354 Repository: ddzipp/AutoAudit

JavaScript 354 stars

jasonacox/TinyLLM

Setup and run a local LLM and Chatbot using consumer grade hardware.

### Overview: TinyLLM is a project that provides a local setup for large language models (LLMs) using consumer-grade hardware, enabling the creation of a small, locally hosted LLM with a ChatGPT-like web interface. ### Key Features: - Supports multiple LLMs (e.g., llama.cpp, vLLM, Ollama) - Builds a local OpenAI API web service - Serves up a Chatbot web interface with customizable prompts, external access, and vector databases ### Tech Stack: Python, CUDA, Docker, and Ollama ### Why it Matters: TinyLLM simplifies the process of setting up and running a local LLM, making it accessible to developers and researchers without the need for expensive hardware.

Python 354 stars

lovit/KR-WordRank

비지도학습 방법으로 한국어 텍스트에서 단어/키워드를 자동으로 추출하는 라이브러리입니다

### Overview: KR-WordRank 是一个基于无监督学习的韩国语文本中自动提取单词和关键词的库。它使用纯 Python 代码实现，无需额外的词形分析器，通过分解句子的词（空格分隔）并构建图来计算每个子串的重要性，然后通过 PageRank 和 HITS 算法中的相似度来确定子串的排名。最终，排名高的子串经过后处理后会被识别为单词。 ### Key features: 1. **无监督学习**: KR-WordRank 是基于无监督学习的方法来提取韩国语文本中的单词和关键词。 2. **纯 Python 实现**: 代码完全由 Python 实现，无需额外的词形分析器。 3. **作者**: Lovit (Hyunjoong Kim)，由 Kim, H. J., Cho, S., & Kang, P. (2014) 提出。 4. **论文引用**: 参考文献 [Kim, H. J., Cho, S., & Kang, P. (2014)]。 5. **CLI**: 提供命令行接口，允许用户通过命令行输入文本文件或 JSON Lines 数据来提取关键词和核心句子。 ### Tech stack: 纯 Python 代码实现。 ### Why it matters: KR-WordRank 是一个强大的工具，特别适用于需要自动提取韩国语文本中关键词和核心句子的应用场景。它简化了关键词提取的过程，使得开发者可以快速地从大量文本数据中提取关键信息，提高了工作效率和

Python 354 stars

lspahija/AIUI

AIUI is a platform enabling seamless two-way verbal communication with AI.

Overview: AIUI is a platform designed for seamless two-way verbal communication with AI models, supporting GPT-4 and GPT-3.5 models, and open models in development. Key features: - Seamless two-way verbal communication with AI models - Supports GPT-4 and GPT-3.5 models - Open models support under development - Built-in language support for English and other languages Tech stack: Python Why it matters: AIUI simplifies interaction with AI models through voice-based communication, making future user interfaces more intuitive and user-friendly. Repository: lspahija/AIUI GitHub stars: 354

Python 354 stars

xr843/fojin

Buddhist Digital Text Platform — 10,500+ texts, 613 sources, trilingual cross-canon, AI Q&A (RAG), knowledge graph, full-text search

### Overview: FoJin is a Buddhist digital text platform that aggregates 612 sources into a searchable, trilingual cross-canon parallel reading platform. It provides answers to questions in the form of AI Q&A, with citations linked back to their source passages. The platform includes a knowledge graph, dictionaries, and a citation drawer for cross-canon comparisons. ### Key features: - **AI Q&A**: Answers retrieved from 612 sources, cross-canon alignment, and clickable citations. - **Cross-Canon Alignment**: 612 sources, 110K+ entities, and 39 dictionaries. - **Knowledge Graph**: Geo map-based knowledge graph. - **Full-Text Search**: 10,500+ texts, 19,000+ volumes in Classical Chinese, Pali, Tibetan, and Sanskrit. - **Cross-Canon Retrieval**: Trilingual cross-canon retrieval with 600+ sources. - **Voice Interface**: Historical Buddhist masters' voices. - **Citation Drawer**: Side-by-side Han, Tibetan, and Sanskrit cross-canon parallels. - **Open, Cross-Canon, Verifiable Buddhist Knowledge Infrastructure**: Corpus for other tools to call. ### Tech stack: - Python ### Why it matters: FoJin provides a comprehensive and verifiable Buddhist knowledge infrastructure, making it easier to understand and verify Buddhist texts. The platform aggregates multiple sources, offers cross-canon alignment, and includes a knowledge graph, dictionaries, and a citation drawer for cross-canon comparisons, making it

Jupyter Notebook 353 stars

FareedKhan-dev/optimize-ai-agent-memory

9 Different Ways to Optimize AI Agent Memories

### Overview: This project aims to explore and evaluate 9 different techniques for optimizing the memory of AI agents. By focusing on various aspects of AI agent memory, including previous context storage, tool calling, database searches, and more, we aim to provide a comprehensive understanding of how to improve the performance and efficiency of AI agents. This project is designed to be practical and scalable, using a simple AI agent as a foundation to observe and analyze the effects of each optimization technique. ### Key Features: - **Previous Context Storage**: Enhances the AI agent's recall of past interactions. - **Tool Calling**: Improves the agent's ability to quickly access and utilize external tools. - **Database Searches**: Enhances the agent's ability to retrieve relevant information from databases. - **Sliding Window Approach**: Optimizes memory usage by selectively retaining the most recent interactions. - **Summarization Based Optimization**: Focuses on summarizing the most important parts of interactions to save memory. - **Retrieval Based Memory**: Utilizes efficient memory management to quickly access and retrieve information. - **Memory Augmented Transformers**: Enhances the performance of transformers by incorporating memory management. - **Hierarchical Optimization for Multi-tasks**: Optimizes memory for handling multiple tasks simultaneously. - **Graph Based Optimization**: Utilizes a knowledge graph to manage and optimize memory. - **Compression & Consolidation Memory**: Reduces the memory footprint by compressing and consolidating data. ### Tech Stack: - Python: The primary language used for development and evaluation. - Jupyter Notebook: A powerful tool for interactive data analysis and

TypeScript 353 stars

KlaatAI/klaatcode

Open-source AI coding agent for the terminal. Claude Code-grade accuracy with smart model routing — uses the right AI model for each task, cutting costs 10x. Supports Claude, GPT, Gemini, DeepSeek & more.

### Overview: Klaat Code is an open-source AI coding agent designed for the terminal. It leverages the Klaatu smart model routing system to provide Claude Code-grade accuracy at a fraction of the cost. The agent routes tasks to the most appropriate model tier based on the complexity of the request, ensuring efficient and cost-effective processing. ### Key Features: - **Smart Model Routing**: Decides the right AI model for each task based on the complexity of the request, optimizing costs. - **Claude Code-Grade Accuracy**: Provides high-quality results with minimal effort. - **Cost Efficiency**: Reduces costs by 10x compared to traditional methods. - **Terminal-First Philosophy**: Runs in the terminal, similar to `gh` on GitHub. - **Code Knowledge Graph**: Uses a semantic search index to query project files efficiently, reducing token usage by up to 5-15x. ### Tech Stack: - TypeScript - Node.js ### Why It Matters: - **Cost Efficiency**: Reduces costs by 10x. - **Claude Code-Grade Accuracy**: Provides high-quality results with minimal effort. - **Smart Model Routing**: Decides the right AI model for each task based on the complexity of the request. - **Terminal-First Philosophy**: Runs in the terminal, similar to `gh` on GitHub. ### Repository: - **Repository**: KlaatAI/klaatcode - **GitHub Stars**: 353 - **GitHub Description**: Open-source AI coding agent for the terminal. Claude Code-grade accuracy with smart model

Python 353 stars

dongyuanxin/news-emotion

📉 金融文本情感分析模型

### Overview: 本项目是金融文本情感分析模型，旨在通过自然语言处理技术对金融领域短文本进行情感分类。模型使用了多种情感词典和机器学习方法，提供了一套完整的文本情感分析框架。 ### Key features: - **情感词典集成**: 包含知网情感词典、台湾大学NTUSD情感词典等多源情感词典。 - **模型训练**: 使用了词频模型、逻辑回归模型等进行模型训练。 - **文本处理**: 支持文本清洗、词向量生成等功能。 - **模型评估**: 提供了留一验证和详细模型参数的展示。 ### Tech stack: Python ### Why it matters: 本项目提供了一套全面的金融文本情感分析解决方案，适用于金融行业进行舆情监控和风险评估。通过模型的准确性和稳定性，可以有效辅助决策者进行市场趋势分析和风险预警。

Jupyter Notebook 353 stars

graviraja/100-Days-of-NLP

Overview: This repository contains a collection of Jupyter notebooks focused on various applications of Natural Language Processing (NLP). The primary language used is Jupyter Notebook, allowing for independent running of code samples. Key features: - Tokenization - Word Embeddings - Word2Vec - Word Embeddings - GloVe - Word Embeddings - ELMo - RNN, LSTM, GRU - Packing Padded Sequences - Attention Mechanism - Luong - Attention Mechanism - Bahdanau - Pointer Network - Transformer - GPT-2 - BERT - Topic Modeling - LDA - Principal Component Analysis (PCA) - Naive Bayes - Data Augmentation Tech stack: Jupyter Notebook Why it matters: This repository provides a comprehensive overview of various NLP techniques and applications, making it valuable for developers looking to deepen their understanding of NLP and its practical applications.

Python 353 stars

tangjiapeng/DiffuScene

[CVPR 2024] DiffuScene: Denoising Diffusion Models for Generative Indoor Scene Synthesis

Overview: DiffuScene is a state-of-the-art diffusion model designed specifically for generating realistic indoor scenes. This innovative approach enables diverse and high-quality scene synthesis, making it a valuable tool for various applications. Key features: - **Scene Synthesis**: DiffuScene can generate diverse and realistic indoor scenes from partial or incomplete scenes. - **Scene Completion**: It facilitates scene completion from partial scenes, enhancing the completeness of generated scenes. - **Scene Arrangement**: The model can arrange objects in a given scene, providing flexibility in scene composition. - **Textual Control**: It supports text-based prompts for generating scenes based on partial scene configurations. Tech stack: The repository utilizes Python, including libraries such as PyTorch, OpenCV, and ChamferDistancePytorch. Why it matters: DiffuScene offers a significant advancement in generative scene synthesis, providing high-quality indoor scene generation that can be applied across various domains, including augmented reality, virtual reality, and content creation.