Projects

### Overview: LayoutParser is a comprehensive toolkit designed for deep learning-based document image analysis. It offers a wide range of tools and APIs to streamline various DIA tasks, including layout detection, OCR, and visualization. The toolkit is optimized for efficient document image analysis and is available as a unified API, making it easy to integrate into various projects. ### Key Features: - **Deep Learning Models**: LayoutParser provides a rich repository of deep learning models for layout detection, including state-of-the-art models like PubLayNet. - **Unified APIs**: It offers a set of unified APIs for easy integration into projects, enabling developers to perform tasks such as layout detection, OCR, and visualization in just a few lines of code. - **Layout Data Structures**: The toolkit comes with a set of layout data structures and APIs that are optimized for efficient document image analysis tasks. - **Visualization**: It provides flexible APIs for visualizing detected layouts, allowing developers to easily identify and analyze document images. - **Community Sharing**: LayoutParser is an open platform that enables the sharing of layout detection models and DIA pipelines among the community, fostering collaboration and innovation. ### Tech Stack: Python ### Why It Matters: LayoutParser simplifies document image analysis by providing a unified toolkit with optimized APIs and a rich repository of deep learning models. Its wide range of features makes it an essential tool for developers working on DIA tasks, enabling them to streamline their workflows and focus on the core functionality of their projects.

Overview: Potpie is a context graph for AI-native software development lifecycle, transforming codebases and SDLCs into living context graphs for AI agents. Key features: - Integrates with various source integrations - Configures agent skills for AI-driven context-based decision-making - Offers a web UI for easy graph exploration Tech stack: Python Why it matters: Potpie transforms traditional software development into a context-based AI-driven development process, enhancing efficiency, accuracy, and project-specific context understanding. Repository: potpie-ai/potpie GitHub stars: 5742 GitHub description: Context Graph for AI Native SDLC README excerpt:

### Overview: UltraRAG is a lightweight RAG development framework based on the Model Context Protocol (MCP) architecture, designed for research exploration and industrial prototyping. ### Key Features: - **Code Efficiency**: UltraRAG requires only a few dozen lines of code to build a high-performance RAG. - **Enhanced Knowledge Ingestion**: Improved knowledge ingestion capabilities. - **Multimodal Support**: Supports multimodal input and output. - **Unified Evaluation System**: A more complete unified evaluation system for RAG models. ### Tech Stack: Python ### Why It Matters: UltraRAG simplifies the process of building complex RAG pipelines, making it easier for researchers and developers to focus on their ideas and innovation without the complexity of "black box" development.

Overview: FlashText is a Python library designed to either extract keywords from sentences or replace keywords in sentences using the FlashText algorithm. Key features: - Extract keywords from sentences - Replace keywords in sentences - Case-sensitive keyword handling - Support for multiple keywords in a dictionary - Span of keywords extraction Tech stack: Python Why it matters: FlashText simplifies keyword extraction and replacement in text, making it useful for natural language processing tasks such as sentiment analysis, text summarization, and more.

### Overview: OpenMed is a local-first healthcare AI platform that leverages Python, MLX, Android, and browser runtimes to extract and de-identify clinical text data. The platform includes 2,200+ medical models, 35 model-backed PII languages, and an Apache-2.0 SDK. It supports 340M+ downloads and 10M+ installations. ### Key Features: - **Local Runtime**: Performs extraction and de-identification after required model artifacts are available. - **Model Catalog**: Access to 35 model-backed PII languages. - **Model Downloads**: Model downloads, remote-provider adapters, telemetry-enabled paths, and user-configured integrations may use a network. - **SDK**: Apache-2.0 SDK source. ### Tech Stack: Python ### Why It Matters: OpenMed is a powerful tool for healthcare professionals to extract and de-identify clinical text data locally, ensuring compliance with HIPAA and other regulations. Its local runtime capabilities make it highly accessible and secure, making it ideal for use on hardware you control.

Overview: SUPIR, or SUPIR Image Restoration, is a cutting-edge project aimed at developing practical algorithms for photo-realistic image restoration in the wild. This project is part of the CVPR 2024 conference and has been featured in the CVPR 2024 conference proceedings and a dedicated online demo at suppixel.ai. Key features: - Practical Algorithms for Photo-Realistic Image Restoration - High Image Quality and Generalization - Online Demo Available at suppixel.ai Tech stack: Python Why it matters: SUPIR offers a robust and efficient solution for image restoration, making it highly relevant for researchers and developers looking to improve image processing techniques in real-world applications. Repository: https://github.com/Fanghua-Yu/SUPIR Primary language: Python GitHub stars: 5688 GitHub description: SUPIR aims at developing Practical Algorithms for Photo-Realistic Image Restoration In the Wild. Our new online demo is also released at suppixel.ai.

Overview: This project is an implementation of a speaker diarization pipeline that combines Whisper ASR capabilities with Voice Activity Detection (VAD) and Speaker Embedding to identify the speaker for each sentence in the transcription generated by Whisper. The pipeline uses Whisper to generate transcriptions, Voice Activity Detection to correct timestamps, and MarbleNet and TitaNet to identify speakers. Key features: - Speaker diarization pipeline using Whisper ASR - Voice Activity Detection (VAD) to correct timestamps - Speaker embedding for accurate transcription - Integration with `ctc-forced-aligner` for time alignment - Use of MarbleNet and TitaNet for VAD and speaker identification Tech stack: Python, Cython, FFMPEG, Cython, Whisper, NeMo, Voice Activity Detection (VAD), Speaker Embedding, MarbleNet, TitaNet, CTC-forced-aligner Why it matters: This project provides a practical solution for automatic speech recognition and speaker diarization, which is valuable in various applications such as transcription, transcription quality improvement, and speaker verification.

### Overview: Baichuan-7B 是一个由百川智能开发的开源可商用的大规模预训练语言模型，基于Transformer结构在1.2万亿tokens上训练，参数量为70亿，支持中英双语，上下文窗口长度为4096。 ### Key features: - 7B参数量 - 13B参数量 - 支持中英双语 - 上下文窗口长度为4096 - 70亿参数量 ### Tech stack: Python ### Why it matters: Baichuan-7B是百川智能开发的一个开源可商用的大规模预训练语言模型，支持中英双语，上下文窗口长度为4096，参数量为70亿，是目前最大的语言模型之一。