Projects

Overview: `clean-text` is a Python package designed to clean and preprocess text data, making it suitable for various applications such as web scraping and social media analysis. Key features: - Fixes various unicode errors - Transliterates text to closest ASCII representation - Lowercases text - Strips line breaks - Replaces code snippets, URLs, email addresses, phone numbers, IP addresses, and file paths - Removes numbers, digits, currency symbols, and punctuations - Preserves specific patterns using regex exceptions Tech stack: Python Why it matters: `clean-text` simplifies the process of text cleaning, making it easier to work with raw user-generated content, ensuring consistent and standardized representations of text data.

### Overview: 全网最全的、持续更新的、最火爆的 100+ 人格技能合集｜多智能体系统 ｜名人/导演/天涯大神/古籍/二次元/职场/情感全品类 ### Key features: - 全网最全的 AI 人格技能合集 - 100+ 种技能，覆盖自我、职场、名人思维、情感陪伴、玄学术数、二次元角色、影视人物、经典IP人设、多角色模拟、Agent 快速人设切换 - 统一规范整理、长期维护更新 - 支持 StepClaw / OpenClaw 全平台 ### Tech stack: Shell ### Why it matters: 为了人类文化传承和延续，这件事值得纯公益做下去。全网最全、100+ AI 人格技能合集，涵盖名人、作家、古籍、二次元、职场、情感等多品类，是目前开源圈内收录最全、更新最稳定、兼容性最好的轻量化人格技能仓库。

### Overview: This curated list of free and low-cost AI tools, LLM APIs, IDEs, agents, and infrastructure for building real AI apps is designed to help developers quickly integrate AI into their projects without breaking the bank. ### Key Features: - **Free GPT-6 / Claude 5.5 / Gemini 3.8 API Access**: Direct access to leading AI models without the need for expensive subscriptions. - **Coding Copilots and AI-Native IDEs**: Tools like Cursor, Trae, and Windsurf that enhance productivity and collaboration. - **Cheapest AI APIs**: Affordable pricing options for various AI services. - **RAG Stack Tools**: Vector databases, embeddings, and frameworks for comprehensive AI capabilities. - **Agent Frameworks and Automation Tools**: Tools for automating tasks and building AI-driven applications. - **Local Models for Privacy**: Secure, local models for privacy-conscious projects. - **Production-Ready Stack Configurations**: Production-ready tools and configurations for seamless integration. - **Claude Sonnet 5.5, Claude Opus 5.5, Haiku 5.5**: High-performance AI models for complex reasoning and autonomous sandboxing. - **OpenCode 167k⭐**: A popular AI copilot with over 167k stars, ensuring reliability and performance. - **Xiaomi MiMo V2.5 Pro**: A high-performance AI copilot with a wide range of features and capabilities. ### Tech Stack: TypeScript ### Why It Matters: This curated list of free AI tools is essential

### Overview: This comprehensive survey explores the state-of-the-art in multimodal chain-of-thought (MCoT) reasoning, providing a detailed taxonomy and analysis of methodologies across various applications. It aims to address the current challenges and guide future research in the field. ### Key features: - **MCoT Reasoning Over Image**: Discusses how MCoT can be applied to reasoning over images. - **MCoT Reasoning Over Video**: Explores the use of MCoT in video reasoning tasks. - **MCoT Reasoning Over 3D**: Examines MCoT in 3D data reasoning scenarios. - **MCoT Reasoning Over Audio and Speech**: Analyzes MCoT in audio and speech reasoning. - **MCoT Reasoning Over Table and Chart**: Discusses reasoning over structured data, specifically tables and charts. - **Cross-modal CoT Reasoning**: Investigates how MCoT can be applied across multiple modalities. ### Tech stack: The survey is presented using LaTeX, focusing on the foundational concepts and definitions of MCoT reasoning. ### Why it matters: This survey fills a gap in the literature by providing a comprehensive overview of the current state of the art in MCoT reasoning, addressing the challenges and future directions in the field.

### Overview: GraphRAG-SDK is a fast and accurate framework for building GenAI applications using GraphRAG, designed to handle large-scale data efficiently and reduce LLM hallucinations. ### Key Features: - **Graph Traversal**: Retrieves connected evidence vector similarity misses, ensuring multi-hop facts are retrieved from the knowledge graph. - **Retrieval Context Traceability**: `MENTIONED_IN` edges trace entity mentions to their source chunks, providing a clear retrieval trail for validation. - **Ontology Constraint**: The graph stores typed, checkable facts instead of free-form model output, ensuring reliability. - **Model Abstention**: Applications can gate generation when evidence is insufficient, preventing hallucinations. ### Tech Stack: Python ### Why It Matters: GraphRAG-SDK simplifies the process of building fast and accurate GenAI applications, reducing LLM hallucinations and providing a reliable framework that can be easily integrated into existing projects.

### Overview: Overcooked-AI is a benchmark environment designed to evaluate the performance of fully cooperative human-AI task coordination. The game simulates the delivery of soups in a ghost town, where agents must place ingredients, wait for the soup to cook, and deliver it efficiently. The environment is used to study human-AI collaboration and has been applied in various research papers, including one on the utility of learning about humans for human-AI coordination. ### Key Features: - **Gameplay Interface:** A web-based interface for playing the game with pre-trained agents. - **Human-AI Data Collection:** Existing human-human and human-AI gameplay data is available for analysis. - **Benchmark Environment:** Based on the popular video game Overcooked, it provides a realistic scenario for evaluating AI performance. - **Open Issues:** Track issues related to the project on GitHub, including bugs and open problems. ### Tech Stack: - **Programming Languages:** Jupyter Notebook - **GitHub Stars:** 1027 ### Why It Matters: Overcooked-AI serves as a valuable benchmark for evaluating the performance of human-AI collaboration in cooperative environments, particularly in the context of video games and AI research. The environment provides a realistic scenario for testing the coordination and decision-making abilities of AI agents, making it an essential tool for researchers in the field of human-AI interaction.

Overview: This guide is designed for high school students looking to explore the field of Machine Learning & Artificial Intelligence. Key features: - Learning the basics of programming in Python - Essential math for Machine Learning, including Calculus, Linear Algebra, Probability, Statistics, and Optimization - A course on the fundamentals of programming in Python - A course on Essential Math for Machine Learning: Python Edition Tech stack: Python, Numpy, Pandas, Matplotlib, Coursera, edX Why it matters: This guide aims to provide a comprehensive learning path for high school students interested in Machine Learning and Artificial Intelligence, without requiring advanced mathematical knowledge beyond what is typically covered in high school.

Overview: Tensorlake is a serverless runtime platform designed for building agentic applications, offering sandbox-native cloud capabilities for AI agents. Key features: - Fastest Filesystem I/O - Fast startup - Snapshots & cloning - Auto suspend/resume - Live migration - Scale up to 5 million sandboxes Tech stack: Rust Why it matters: Tensorlake simplifies the development and deployment of AI agents, providing a serverless and sandbox-native cloud environment that accelerates the development process and reduces operational overhead.

### Overview: This repository serves as a comprehensive collection of synthetic datasets and tools for computer vision research. It aims to provide researchers with a diverse set of datasets and tools to enhance their studies and experiments. ### Key Features: - **Synthetic Image Datasets**: Includes datasets like SunCG, Minos, House3d, Procedural Human Action Videos, SURREAL, Virtual KITTI, Synthia, Sintel, SceneFlow, and 4D Light Fields. - **3D Model Repository**: Features efforts to create realistic 3D model repositories such as ShapeNet, 3dscan, and seeing3Dchairs. - **Tools**: Offers tools like AIPlayground, AirSim, CARLA, Unity ML agents, UETorch, UnrealCV, VizDoom, OpenAI Universe, Blender addon for 4D light field rendering, Event-Camera Dataset and Simulator, and NVIDIA Deep learning Dataset Synthesizer. ### Tech Stack: The repository is built using Python, leveraging Unreal Engine 4 (UE4) as the primary development environment. ### Why It Matters: This repository is crucial for researchers and practitioners in computer vision, providing a rich set of synthetic data and tools that can significantly accelerate research and development in the field.

### Overview: rclip is a powerful, command-line tool designed for searching images within your local filesystem. It enables users to find photos by describing them, using natural language queries, or by providing an example image. The tool supports reverse image searches and allows for combined and weighted queries, making it highly versatile and efficient. ### Key features: - **Natural-Language Search**: Find photos by describing them, e.g., `rclip "two parrots on a branch"`. - **Reverse Search**: Search by an example image from a local path or a URL. - **Combined Queries**: Mix and weight text and image queries, e.g., `rclip "2:golden retriever" + "./pool.jpg" - fruit`. - **Offline Operation**: Works fully offline, ensuring your photos never leave your computer. - **Wide Format Support**: Supports multiple image formats, including `jpg`, `png`, `webp`, `tiff`, `gif`, and more, with native HEIC on macOS/Windows. - **Fast Incremental Indexing**: Only new and changed images are reprocessed on subsequent runs, ensuring efficient performance. - **Interactive Terminal UI**: Search a responsive thumbnail grid without leaving the terminal. - **Terminal Previews**: View images inline in terminals supporting Kitty graphics, such as Kitty, Ghostty, latest iTerm2, and Konsole. - **Cross-Platform Compatibility**: Works fully on Linux, macOS (Apple Silicon), and Windows. ### Tech Stack: Python ### Why it Matters: rclip is a valuable tool for developers and photographers

Overview: Cloudroom core is an open-source, self-hostable Rust runtime designed to run Claude Code, Codex, Pi, OpenCode, Hermes, Cursor, and Devin agents on your own Linux machine. It powers the cloud side of the Cloudroom app, which runs on your Mac or in the cloud. Key features: - Runs agents on your own machine - Saves chat history locally - Works with Claude Code, Codex, Pi, OpenCode, Hermes, Cursor, and Devin - Agents use your own logins - No need for Cloudroom account - One Rust service per machine - No database needed Tech stack: Rust Why it matters: Cloudroom core is essential for developers who want to run their own agents and chat history without relying on a centralized service. It provides a self-hosted solution for those who prefer to manage their own infrastructure, ensuring complete control and privacy. Repository: davidondrej/cloudroom-core GitHub stars: 980 GitHub description: Open-source, self-hostable Rust runtime that runs Claude Code, Codex, and Pi in the cloud.

### Overview: CLUECorpus2020 是一个包含100GB高质量中文预训练语料的项目，主要用于中文预训练模型、语言模型和生成型任务。该项目通过清洗Common Crawl的中文部分，提供了丰富的语料资源，涵盖了简体中文、传统中文、英文、日语和韩语等多种语言。 ### Key features: - **100GB 大规模中文语料库**：包含100GB高质量中文预训练语料。 - **预训练模型**：提供高质量中文预训练模型，包括BERT-base、BERT-base mm、小词表版本、相似度预训练模型等。 - **小词表版本**：发布专用于简体中文NLP任务的小词表，方便快速训练和应用。 - **数据格式**：语料库支持预训练格式，包含多个文件夹和小文件，格式符合预训练模型要求。 ### Tech stack: - **GitHub**：项目托管和代码管理。 ### Why it matters: - **广泛适用性**：适用于多种中文NLP任务，包括预训练、语言模型和生成型任务。 - **高质量数据**：提供100GB的高质量中文语料，有助于提升模型性能。 - **方便获取**：语料库数据量大，便于直接用于训练和测试。 ### Repository: Repository: CLUEbenchmark/CLUECorpus2020 GitHub Stars: 1023 ### Data: - **100GB 大

### Overview: A curated list of resources dedicated to Japanese Natural Language Processing (NLP), including Python libraries, LLMs, dictionaries, corpora, and datasets. The repository features Claude Code and Codex skills to search and discover relevant resources. ### Key Features: - Python libraries - LLMs (Large Language Models) - Dictionaries - Corpora - Datasets ### Tech Stack: - Python ### Why It Matters: This repository serves as a comprehensive resource for developers interested in Japanese NLP, providing access to various tools and datasets that can be used for tasks such as text generation, translation, and more. The inclusion of Claude Code and Codex skills allows for direct integration of these tools into the development workflow, enhancing efficiency and functionality.

### Overview: FlowEdit is an official implementation of the paper "FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models". This project leverages pre-trained flow models to achieve zero-shot image editing, making it a valuable tool for text-based image manipulation. ### Key Features: - **Text-based Image Editing**: FlowEdit allows users to edit images by providing a source prompt and target prompts, with the system generating the necessary edits. - **Pre-trained Models**: Utilizes pre-trained flow models to perform the editing process, ensuring high-quality results without the need for additional training. - **Versatile Usage**: Supports both Stable Diffusion 3 and Flux models, making it compatible with various downstream tasks. ### Tech Stack: - **Python**: The primary programming language used for the implementation. - **PyTorch**: Used for the development and training of the flow models. ### Why it Matters: - **Efficiency**: FlowEdit simplifies the process of image editing by automating the generation of edits based on prompts, reducing the need for manual intervention. - **Versatility**: The project supports multiple models, including Stable Diffusion 3 and Flux, making it a versatile tool for a wide range of applications. - **Zero-Shot Capabilities**: The ability to edit images without the need for additional training makes it particularly useful for tasks where training data is limited. ### Installation: - Clone the repository: `git clone https://github.com/fallenshock/FlowEdit.git` - Install required dependencies: `pip install torch diffusers transformers accelerate sentencepiece protobuf

Overview: This repository serves as a central hub for documentation, SLURM scripts, logs, and compute environment information related to the engineering/scaling working group. It includes detailed information about the team's work environment, experiments, datasets, and training sessions. Key features: - Comprehensive documentation and scripts for various experiments and training sessions. - Detailed chronicles of trainings, including scripts, logs, and tensorboard summaries. - Access to SLURM scripts and logs for efficient computing environment management. Tech stack: Shell-based scripts and documentation. Why it matters: This repository provides a crucial resource for the engineering/scaling working group, offering insights into the team's work environment, experiments, and training sessions, which is essential for evaluating and planning future projects. GitHub stars: 1020 Repository: bigscience-workshop/bigscience

Overview: BlenderNeRF simplifies the creation of synthetic NeRF and Gaussian Splatting datasets within Blender, making it easy and fast for artists, researchers, and enthusiasts alike. Key features: - Easy NeRF and Gaussian Splatting dataset creation - Single-click setup for rendering and camera parameters - Full user control over 3D scenes and cameras - Training and testing data stored in a single ZIP file Tech stack: Python Why it matters: It streamlines the process of generating high-quality synthetic data for NeRF and Gaussian Splatting, allowing users to focus on artistic expression or research without the complexity of traditional setup. Repository: maximeraafat/BlenderNeRF GitHub stars: 1019

### Overview: PythonSIFT is a clean and concise Python implementation of SIFT (Scale-Invariant Feature Transform), designed to be a drop-in replacement for OpenCV's SIFT. This project aims to provide a detailed understanding of the SIFT algorithm and its implementation, making it accessible for computer vision enthusiasts. ### Key features: - **Python Implementation**: Pure Python with NumPy support. - **OpenCV Compatibility**: Returns OpenCV `KeyPoint` objects and descriptors, compatible with OpenCV's SIFT. - **Educational Tool**: Ideal for learning the details behind SIFT and its implementation. ### Tech stack: - Python 3 - NumPy - OpenCV-Python ### Why it matters: PythonSIFT is an educational tool that helps beginners understand the intricacies of SIFT, making it a valuable resource for anyone interested in computer vision and image processing. ### Repository: rmislam/PythonSIFT - **GitHub Stars**: 1018 - **Primary Language**: Python - **Dependencies**: Python 3, NumPy, OpenCV-Python ### Usage: ```python import cv2 import pysift image = cv2.imread('your_image.png', 0) keypoints, descriptors = pysift.computeKeypointsAndDescriptors(image) ``` ### Dependencies: - Python 3 - NumPy - OpenCV-Python ### Tutorial: - Detailed explanation of the SIFT implementation, including a step-by-step guide. ### Template Matching Demo: - Adapted OpenCV's SIFT template matching demo

### Overview: Codex CLI is a powerful tool designed to streamline and enhance your development workflow. Built on the OpenAI Codex, it offers a seamless integration of AI capabilities directly into your command-line interface, making it easier to manage and deploy AI models. ### Key Features: - **Commands**: Built-in slash commands for session control, including `/plan`, `/fast`, `/fork`, `/review`, `/status`, `/mcp`, `/agent`, `/apps`, `/model`, and `/permissions`. - **Subagents**: Custom agents registered under `[agents.<name>]` with dedicated TOML role configs, parallel subagent orchestration, and CSV batch processing. Global settings live under `[agents]` (`max_threads`, `max_depth`, `job_max_runtime_seconds`). - **Skills**: Reusable instruction packages with required `name` + `description` metadata and progressive disclosure via `scripts/`, `references/`, `assets/`, and optional `agents/openai.yaml`. - **Plugins**: Distributable bundles combining skills + app integrations + MCP servers — local/personal marketplace system. - **Marketplace**: Plugin catalog system (v0.121.0+) — `codex plugin marketplace add\|upgrade\|remove` accepts GitHub shorthand, git URLs, and local directories. - **Memories**: Cross-session memory pipeline (v0.119.0+) — enable via `[features] memories = true` and configure under `[memories]`. ### Tech Stack: Python ### Why It Matters: Codex CLI is essential for anyone

### Overview: TensorFlowASR is an open-source project that implements advanced Automatic Speech Recognition (ASR) models using TensorFlow 2. It supports various languages and models such as DeepSpeech2, Jasper, RNN Transducer, ContextNet, and Conformer. These models can be converted to TFLite for efficient deployment. ### Key features: - **Models Supported**: Conformer, ContextNet, Jasper, RNN Transducer, DeepSpeech2, and many more. - **Models Conversion**: Models can be converted to TFLite for reduced memory and computation. - **Languages Supported**: Supports characters and subwords in multiple languages including English and Vietnamese. ### Tech stack: Python 3.12 or 3.13 with optional NVIDIA GPU and development tooling. ### Why it matters: TensorFlowASR provides a state-of-the-art ASR solution, making it an essential tool for developers working on speech recognition projects. Its flexibility and support for various models make it a valuable resource for anyone looking to improve their ASR system.

Overview: SPLADE is a neural retrieval model that learns query/document sparse expansion using the BERT masked language model (MLM) head and sparse regularization. This model is designed to improve efficiency and effectiveness in both in-domain and out-of-domain evaluations, particularly on the BEIR benchmark. Key features: - Training code for SPLADE and rerankers training. - Static pruning of SPLADE indexes for reproducibility. - New models coming soon on GitHub. - Code for training with several negatives using HF Trainer. - Weights for models trained under various settings available on Naver Labs Europe website and Hugging Face. Tech stack: Python. Why it matters: SPLADE models are effective and efficient, benefiting from sparse representations that are better at generalizing on out-of-domain data, such as the BEIR benchmark. Repository: naver/splade Primary language: Python GitHub stars: 1017 GitHub description: SPLADE: sparse neural search (SIGIR21, SIGIR22)