Projects

### Overview: LangChain4j-AIDEEPIN is an AI application platform designed to rapidly build intelligent business assistants. It integrates various AI functionalities such as chat, knowledge base, workflow orchestration, long/short-term memory, and more, making it a versatile tool for developers. ### Key Features: - **AI Chat**: Multi-character (multi-conversation) with configurable prompts, models, and parameters. Supports streaming output. - **Image Generation**: Text-to-image and image editing. Supports GPT-Image-2, DashScope Wanx, and more. - **Knowledge Base (RAG)**: Supports both vector search and knowledge graph retrieval. - **AI Workflow**: Visual editor with conditional branching, parallel execution, and built-in nodes for LLM calls, knowledge base queries, human feedback, and more. - **MCP Service Marketplace**: Integrate MCP services to extend AI with external tools and data sources. - **ASR & TTS**: Flexible text/voice input and output combinations (text⇄text, text⇄voice, voice⇄text, voice⇄voice), with selectable AI voice tones. - **Short & Long-term Memory**: Automatically extracts and stores key information from conversations for personalized responses based on historical context. - **Storage**: Local file storage, Alibaba Cloud OSS integration. - **Open API**: RESTful API for Characters, Knowledge Bases, and Workflows — generate API keys and integrate with external systems. Supports streaming and blocking response modes. ### Tech Stack: Java ### Why it Matters: LangChain4j-AIDEEPIN provides a comprehensive solution

### Overview Xtreme1 is an all-in-one open-source platform designed for multimodal training data annotation and management. It supports 3D LiDAR point cloud, image, and LLM datasets, providing AI-fueled tools for enhancing annotation efficiency and model training. ### Key Features - **Image Annotation**: Supports bounding boxes, polygons, polylines, and key points. - **Lidar-Camera Fusion Annotation**: Integrates with OpenPCDet and AB3DMOT for 3D LiDAR and sensor fusion. - **Built-in Pre-labeling Models**: Includes YOLOR and OpenTSNE for 2D/3D object detection. - **Configurable Ontology Center**: Manages general classes and attributes for model training. - **Data Management and Quality Monitoring**: Ensures data integrity and error detection. - **Model Results Visualization**: Helps in evaluating model performance. - **RLHF for Large Language Models**: Beta version for large language model training. ### Tech Stack - TypeScript ### Why It Matters Xtreme1 simplifies the annotation process, making it more efficient and effective for tackling machine learning challenges in computer vision and LLM. Its AI-powered tools enhance your annotation capabilities, powering your projects with unprecedented accuracy and speed. ### Repository - Repository: xtreme1-io/xtreme1 - GitHub Stars: 1385 - GitHub Description: Xtreme1 is an all-in-one open-source platform for multimodal training data. ### Installation - **Prerequisites**: - Operating System: Any OS

### Overview: KnowLM is an Open-sourced Knowledgable Large Language Model Framework, designed for pre-training, fine-tuning, and knowledge augmentation of large language models. It includes a model zoo with models like ZhiXi and OneKE, and supports instruction processing and knowledge editing. ### Key Features: - **Model Zoo**: Accessible models like ZhiXi and OneKE for quick implementation. - **Model Pre-Training and Fine-Tuning**: Standard framework for LLM pre-training and fine-tuning. - **Model Zeroshot**: Models are available for zero-shot learning. - **Knowledge Augmentation**: Knowledge augmentation using RAG. - **Instruction Processing**: EasyInstruct-based instruction processing module. - **Model Deployment**: Inference and deployment capabilities. - **Regular Updates**: Regularly updated weights and datasets to support new features and models. ### Tech Stack: Python ### Why it Matters: KnowLM provides a comprehensive framework for large language models, making it easier to implement knowledge extraction and information processing tasks. Its model zoo and instruction processing capabilities make it a valuable tool for researchers and developers in the field of natural language processing. ### Repository: zjunlp/KnowLM ### GitHub Stars: 1385 ### Primary Language: Python ### GitHub Description: An Open-sourced Knowledgable Large Language Model Framework.

### Overview: Aegis is a method pack designed to make AI coding agents trustworthy and reliable. It ensures fewer reworks, safer changes, and proof before "done" by aligning with your project's real baseline, freezing a held-out benchmark, and leaving simple tasks alone. ### Key Features: - **Fewer Reworks:** Your agent and your project's real baseline align before touching code, reducing rework. - **Safer Changes:** A 61.67% contract pass rate compared to 93.33% after A/B benchmark. - **Proof Before "Done":** Completion claims are accompanied by fresh verification evidence, ensuring safety. - **No Ghost Code:** Retired fallbacks and old paths are tracked or removed, preventing ghost code. ### Tech Stack: Python ### Why It Matters: Aegis ensures your AI coding agents are more reliable and trustworthy, reducing rework, improving safety, and providing clear evidence of completion.

Overview: Memory OS is a 7-layer memory operating system designed for the Hermes Agent, providing persistent memory, structured facts, fabric recall, auto-curated wiki, and context injection. It runs locally and integrates seamlessly with any LLM provider, offering a real-time, long-term collaborator experience. Key features: - Persistent memory - Structured facts with trust scoring - Self-curating wiki pipeline - Semantic search across every conversation you've ever had Tech stack: Python Why it matters: Memory OS turns Hermes Agent into a reliable, long-term collaborator that remembers projects, decisions, reasoning, and context, enhancing productivity and decision-making. Repository: ClaudioDrews/memory-os GitHub stars: 1380 GitHub description: A 7-layer memory operating system for Hermes Agent — persistent memory with Qdrant, structured facts, fabric recall, auto-curated wiki, and surgical context injection. Runs locally, any LLM provider.

Overview: This project provides a collection of skills for Claude Code, including before-and-after screenshots, service layer architecture guidance, evidence-driven testing, Greptile feedback handling, and more. These skills are designed to streamline workflows and ensure that changes are thoroughly tested and documented. Key features: - **before-and-after**: Captures before/after screenshots and generates a PR-ready markdown comparison table for visual proof. - **code-structure**: Guides the service layer architecture for projects without established conventions. - **evidence-driven-testing**: Records visual proof while testing UI behavior and posts evidence to PRs and trackers. - **greploop**: Addresses Greptile feedback on PRs, MRs, or shelved changelists. - **greploop-apps**: A helper script for handling Greptile feedback on apps. Tech stack: Python Why it matters: These skills are essential for Claude Code developers to streamline their workflows, ensure thorough testing, and maintain a clean and documented codebase.

### Overview: This repository is a curated catalog of Meta Muse integrations and community connector skills, providing a comprehensive source-backed overview of available tools and their capabilities. ### Key Features: - Integrations from Meta-named platforms - 150 community connector skills, sourced from an MIT-licensed project - Detailed descriptions, authentication notes, and maturity status for each skill - Information on built-in Meta connectors and their source links - A community-maintained catalog of integrations and skills for Meta Muse ### Tech Stack: Python ### Why It Matters: This repository serves as a valuable resource for developers looking to integrate Meta Muse with various applications and tools. It provides a curated list of skills and integrations, ensuring that users can easily find and utilize the right tools for their projects. ### Repository: Anil-matcha/awesome-muse-connectors Primary Language: Python GitHub Stars: 1379

### Overview: FireRed-Image-Edit is a powerful image editing foundation model achieving open-source state-of-the-art performance with precise instruction following, high-fidelity generation, superior identity consistency, and seamless multi-element fusion. ### Key features: - **Strong Editing Performance**: Achieves leading open-source results with accurate instruction following, high image quality, and consistent visual coherence. - **Identity Consistency**: Ensures precise and consistent editing across various image elements. - **Multi-Element Fusion**: Seamlessly merges multiple image elements into a cohesive final image. - **Stylized Text Reference**: Enhances the realism and authenticity of text references in the edited images. ### Tech stack: Python ### Why it matters: FireRed-Image-Edit is a foundational model for image editing that offers high performance and consistent results, making it valuable for both professional and personal use cases.

### Overview: Skills is a comprehensive library of reusable AI agent skills designed for software factories. It enables developers to explore ideas, write specifications, implement, review, and run autonomous research, all while maintaining a living document for continuous improvement. ### Key Features: - **Exploration**: `/explore-unknowns` for mapping the fog of war. - **Specification Writing**: `/write-spec` for transcription of decisions. - **Implementation**: `/goal` for kicking off the loop with a goal. - **Review**: `/review-choices` for consolidating and reviewing the choices made during the implementation. ### Tech Stack: Skills is built using HTML and JavaScript, leveraging the Claude Code, Codex, and other skill-compatible agents. ### Why It Matters: Software is moving from tasks to factories, where agents pursue goals autonomously until the output can be trusted. These skills enable developers to break down complex tasks into independently verifiable pieces, ensuring that each piece proves its correctness before moving on to the next. This approach helps maintain a living document for continuous improvement, ensuring that the goal is completed efficiently and reliably.

### Overview: "结巴"中文分词：一款高质量的PHP中文分词、中文断词组件。它基于 Trie 樹結構實現了高效的詞圖掃描，支持三種分詞模式：默認精確模式、全模式和搜尋引擎模式。此外，它還支持繁體斷詞、自定義詞典、多語言CJK文本處理（中文、日文、韓文）和TF-IDF分詞整合和詞性標註。目前已經為一個獨立分支，歡迎有興趣的開發者一起加入開發。 ### Key features: - 支持三種分詞模式：默認精確模式、全模式和搜尋引擎模式。 - 支持繁體斷詞。 - 支持自定義詞典。 - 支持多語言CJK文本處理（中文、日文、韓文）。 - 支持 TF-IDF 分詞整合和詞性標註。 - 支持記憶體管理和快取優化。 - 支持自定義詞性標籤。 ### Tech stack: PHP ### Why it matters: "结巴"中文分词是一款高质量的PHP中文分词、中文断词组件。它基于 Trie 樹結構實現了高效的詞圖掃描，支持三種分詞模式：默認精確模式、全模式和搜尋引擎模式。此外，它還支持繁體斷詞、自定義詞典、多語言CJK文本處理（中文、日

Overview: AllenAI's Objaverse-XL is a comprehensive dataset of over 10 million 3D objects, designed to enhance the training of 3D models. Key features: - Contains scripts for downloading and processing the dataset. - Provides APIs for downloading and processing the dataset. - Offers a large-scale dataset that is over an order of magnitude larger and more diverse than the original Zero123 model. - Improves 0-shot generalization abilities, allowing Zero123-XL to perform novel view synthesis on sketches, cartoons, and people. - Enables image → 3D and text → 3D transformations using the base Zero123-XL foundation model. - Demonstrates strong scaling trends for both Zero123-XL and PixelNeRF. - Provides Blender rendering scripts for further visualization and exploration. - Offers a variety of licenses for the dataset, including the ODC-By v1.0 license for the whole dataset and different licenses for individual objects. Tech stack: Python Why it matters: The large-scale dataset of over 10 million 3D objects in Objaverse-XL is crucial for training and improving the generalization abilities of 3D models, particularly in zero-shot scenarios, making it a valuable resource for researchers and developers in the field of 3D object recognition and rendering.

Overview: This Jupyter book serves as a comprehensive guide to physics-based deep learning, focusing on practical application through interactive Jupyter notebooks. It covers advanced techniques such as probabilistic surrogate models, differentiable simulations, and reinforcement learning, aiming to redefine computational science. Key features: - Hands-on, practical guide to physics-based deep learning - Interactive Jupyter notebooks for quick application - Focus on probabilistic models, differentiable simulations, and reinforcement learning - Integration with existing knowledge of physics - Use of neural networks to solve PDE problems, combine with numerical simulations, and choose the right architecture Tech stack: Jupyter Notebook Why it matters: This book aims to bridge the gap between data-driven approaches and physical simulations, utilizing neural networks to enhance computational science, particularly in scenarios where numerical methods are repeatedly applied. Repository: tum-pbs/pbdl-book Primary language: Jupyter Notebook GitHub stars: 1377 GitHub description: Welcome to the Physics-based Deep Learning Book v0.3 - the GenAI Edition

Overview: This repository is dedicated to a comprehensive collection of research papers, benchmarks, and open-source projects related to large language model-based text-to-SQL (LLM-based Text-to-SQL). It serves as a curated list of resources from our survey paper, "Next-Generation Database Interfaces: A Survey of LLM-based Text-to-SQL," and will continuously update to reflect the latest advancements and notable contributions in the field. Key features: - A curated list of research papers, benchmarks, and open-source projects on LLM-based Text-to-SQL. - All contents from our survey paper, "Next-Generation Database Interfaces: A Survey of LLM-based Text-to-SQL." - Continuous updates to incorporate the latest advances and notable contributions in the field. - A taxonomy of in-context learning and fine-tuning methods. - Post-annotated datasets for better evaluation and benchmarking. - Datasets including original and post-annotated versions for diverse applications. - Projects showcasing real-world implementations and use cases. - A citation guide for referencing our survey and repository. Tech stack: Python, Markdown, Git, and various open-source tools for data collection and organization. Why it matters: This repository is essential for researchers, developers, and practitioners interested in the field of text-to-SQL, providing a valuable resource for staying updated on the latest advancements and best practices in the field.

Overview: GeneticSharp is a fast, extensible, multi-platform, and multithreading C# library designed to simplify the development of applications using Genetic Algorithms (GAs). Key features: - Fast execution speed - Extensive feature set - Support for .NET 6, .NET Standard, and .NET Framework - Cross-platform compatibility - Multithreading capabilities Tech stack: C# Why it matters: GeneticSharp simplifies the implementation of Genetic Algorithms in .NET applications, making it easier to develop applications that leverage genetic optimization techniques.

### Overview: NGT is a high-speed approximate nearest neighbor search library for high-dimensional vector spaces, supporting graph and tree-based methods. It is designed for efficient indexing and searching of large datasets in multiple dimensions. ### Key Features: - **Graph and Tree-Based Method (NGT)**: Performs searches using a graph and tree structure. - **Quantized Graph-Based Method (QG)**: Utilizes quantized graph for faster search times. - **Quantized Blob Graph-Based Method (QBG)**: Provides a more efficient search method for high-dimensional data. - **Scalar Quantization**: Available for both graph and blob indices. - **FP16 Support**: Supports half-precision floating point operations. - **Command Line Interface**: Provides a command-line interface for easy usage. ### Tech Stack: C++ ### Why it Matters: NGT is essential for applications requiring high-speed approximate nearest neighbor searches in high-dimensional data, making it suitable for various fields such as recommendation systems, image retrieval, and more. ### Repository: NGT-labs/NGT ### GitHub Stars: 1375 ### Installation: - **Linux**: Use `cmake` and `make` with appropriate flags. - **CentOS**: Install BLAS and LAPACK libraries. - **Ubuntu**: Use `brew` to install BLAS and LAPACK. - **MacOS**: Use `homebrew` and `MSYS2 / UCRT64` for MinGW-w64 GCC. - **Windows**: Use `MSVC` for native support, requiring `-DNG

Overview: This repository contains foundational resources for software engineers to build a comprehensive understanding of AI, including machine learning, large language models, and engineering aspects. Key features: - General resources for software design and technical judgment - ML Foundations: Hands-On Machine Learning with Scikit-Learn and PyTorch - LLMs: Neural Networks & Transformer Series, Build a Large Language Model (From Scratch), Build a Reasoning Model (From Scratch) Tech stack: Python, Jupyter Notebook, Git Why it matters: This guide is essential for software engineers looking to build a solid foundation in AI, machine learning, and large language models, providing practical insights and hands-on experience.

### Overview: This repository contains a collection of NLP tutorials built on PyTorch, focusing on various tasks such as text classification, neural machine translation, and natural language understanding. ### Key features: - Text Classification: Implementations of Text Classification using CBoW, LSTM, fastText, and Transformer models. - Neural Machine Translation: Training seq2seq and Transformer models for translation tasks. - Natural Language Understanding: Implementing a simple Neural Language Model for natural language tasks. ### Tech stack: PyTorch ### Why it matters: These tutorials provide practical examples and step-by-step guides for implementing common NLP tasks, helping developers and learners to understand and apply these techniques effectively.

Overview: This Python client library provides a simple and efficient interface to interact with the Qdrant vector search engine, allowing developers to perform vector search and other operations directly from Python. Key features: - Type hints for all API methods - Local mode for development and testing - Support for both REST and gRPC protocols - Minimal dependencies - Extensive test coverage Tech stack: Python Why it matters: This client library simplifies working with Qdrant, making it easier for developers to integrate vector search capabilities into their applications, especially for local development and testing purposes. Repository: qdrant/qdrant-client GitHub stars: 1374

Overview: Zemberek-NLP is a comprehensive Natural Language Processing (NLP) toolkit for the Turkish language, providing tools for morphological analysis, tokenization, normalization, named entity recognition, text classification, language identification, language modeling, and more. Key features: - Morphology - Tokenization - Normalization - Named Entity Recognition (NER) - Text Classification - Language Identification - Language Modeling - Console Applications - gRPC Server Tech stack: Java Why it matters: Zemberek-NLP offers a robust set of NLP tools for Turkish developers, enabling them to process and analyze text data efficiently and effectively.

### Overview: TorchSSL is a PyTorch-based toolbox for semi-supervised learning, featuring 9 popular algorithms and 5 supported datasets. It is an upgrade of the deprecated TorchSSL, with improved flexibility and extensibility. ### Key Features: - **Supported Algorithms**: PiModel, MeanTeacher, PseudoLabel, VAT, MixMatch, UDA, ReMixMatch, FlexMatch, FreeMatch, and SoftMatch. - **Supported Datasets**: CIFAR-10, CIFAR-100, STL-10, SVHN, and ImageNet. - **Training Efficiency**: TorchSSL reduces training time by 12.5% compared to using the deprecated TorchSSL. ### Tech Stack: Python ### Why it Matters: TorchSSL provides a comprehensive and flexible toolkit for semi-supervised learning, enabling fair and efficient comparison of various algorithms and datasets.