Projects

### Overview: bkit is a Claude Code plugin designed to verify AI-generated code against its own design specifications, ensuring robust and production-quality software. It simplifies the process for developers, even those new to AI-based coding, by providing context and quality gates to ensure the right context is always available. ### Key Features: - **Auto-Repair**: AI proposes, bkit measures the result against its own design spec, and auto-repairs the gap when it drifts. - **Context Engineering System**: 44 skills, 34 specialist agents, 11 quality gates, and a memory that survives across sessions. - **3 Commands**: Anyone, even a first-time vibe-coder, can ship robust software with bkit. - **Claude Code v2.1.143 or Later**: Requires the latest version of Claude Code to use bkit. - **8-Language Auto-Trigger**: Supports 8 languages, auto-detecting commands and translating them. ### Tech Stack: - JavaScript ### Why It Matters: bkit streamlines the development process, making it accessible to developers of all skill levels. It ensures high-quality software delivery without the need for senior engineer intuition, providing a safety net and a team-in-a-box approach. The context-engineering system and quality gates help prevent drift and ensure context is always available, making long projects manageable and reducing the risk of losing context. ### Repository: ww-w-ai/bkit-claude-code ### GitHub Stars: 604 ### GitHub Description: A Claude Code plugin that verifies

Overview: UniEmployee 是一款面向企业的数字员工构建与运行平台，通过固化专业员工的工作经验、业务流程和判断标准，将其转化为可配置、可审批、可观测的AI数字员工，帮助企业提升生产力。 Key features: - **数字员工构建与管理**: 页面化配置，支持员工分配、软删除与恢复，内置8个示例员工。 - **流程型技能与SOP**: 技能以规程沉淀，播种进Store供模型按需查阅，关键业务流程用StateGraph固化。 - **数据分析与报告工作台**: 独立的`xiaoshu`工作台，支持多种数据源，分析报告与市场简报可下载。 - **企业知识与本体**: 接入FAQ、产品Wiki、RAGFlow等多源知识与结构化企业本体，支持实体/关系管理。 - **连接器与工具生态**: 支持MCP接入外部系统，内置工单、搜索、知识库、文档生成等原子工具。 - **跨会话长期记忆**: 按`(user_id, employee_id)`隔离，落盘Store数据库，重启不丢。 - **超长对话自动压缩**: 基于deepagents内置`SummarizationMiddleware`，对话自动折叠为摘要。 - **全链路可观测**: 记录对话输入输出、耗时、token消耗，可回放定位。 - **HITL + 多层安全护栏**: 高风险动作触发流程中途中断，等待

### Overview: Welcome to the curated list of AI tools, models, and platforms designed for developers, professionals, and enthusiasts. This repository, updated annually, serves as a comprehensive resource for exploring the rapidly evolving AI landscape. It includes systems across large language models, multimodal generation, AI agents and automation, developer platforms, and infrastructure tools. ### Key Features: - **Large Language Models (LLMs):** Tools and platforms for advanced natural language processing and AI-driven communication. - **Multimodal Generation:** Systems for generating images, videos, and audio from text inputs, enhancing creativity and content creation. - **AI Agents and Automation:** Platforms and tools for automating tasks, improving efficiency, and enhancing productivity. - **Developer Platforms and APIs:** Tools and platforms for developers to build and deploy AI applications, including frameworks and libraries. - **Infrastructure and Supporting Tools:** Resources for setting up and managing AI infrastructure, including cloud services and development environments. ### Tech Stack: The repository uses Markdown for documentation and GitHub for hosting, ensuring a clean and accessible format. ### Why it Matters: This repository is a valuable resource for anyone looking to understand and utilize AI tools and platforms. It serves as a go-to source for developers, professionals, and enthusiasts seeking to explore the latest AI tools and platforms.

Overview: This project involves the creation of a simple chatbot using Python and the Natural Language Toolkit (NLTK) library. The chatbot is designed to be a basic conversational interface, utilizing NLTK's capabilities to analyze and generate text. Key features: - Utilizing NLTK for text processing and analysis - Implementing a basic conversational interface Tech stack: Python, NLTK Why it matters: This project serves as an introduction to NLP and chatbots, allowing developers to practice and understand the fundamentals of natural language processing and building conversational applications.

### Overview: OpenBrowserClaw is a browser-native personal AI assistant that runs entirely in a browser tab, eliminating the need for any infrastructure. ### Key Features: - **Zero Infrastructure**: The browser acts as the server, eliminating the need for any external servers or infrastructure. - **NanoClaw Philosophy**: Built with the same principles as NanoClaw, focusing on simplicity and understanding. - **Anthropic API Integration**: Orchestrates interactions with the Anthropic API, allowing for seamless integration with the Claude model. ### Tech Stack: TypeScript ### Why It Matters: OpenBrowserClaw is a browser-native personal AI assistant that runs entirely in a browser tab, eliminating the need for any infrastructure. This project is designed to be safe and secure, as it is not affiliated with any cryptocurrency, meme coin, token, or social media account. Stay safe and do your own research. ### Repository: wexare-ai/openbrowserclaw ### GitHub Stars: 604 ### GitHub Description: Browser-native personal AI assistant. Zero infrastructure — the browser is the server.

### Overview: NVIDIA Deep learning Dataset Synthesizer (NDDS) is a UE4 plugin designed to empower computer vision researchers by exporting high-quality synthetic images with metadata. It supports various image types and components for generating randomized scenes, including lighting, objects, camera position, poses, textures, and distractors. The plugin is essential for creating randomized training data for deep neural networks. ### Key features: - **Exporting Synthetic Images**: NDDS allows researchers to export high-quality synthetic images with metadata. - **Randomization Components**: Includes lighting, objects, camera position, poses, textures, and distractors for generating randomized scenes. - **Metadata Support**: Supports images, segmentation, depth, object pose, bounding box, keypoints, and custom stencils. ### Tech stack: C++ ### Why it matters: NDDS is crucial for training and testing deep learning systems by providing synthetic data that can be used for real-world applications, overcoming the limitations of hand-labeled data. It enables successful domain transfer and showcases the potential of simulators in computer vision research.

Overview: hcom is a powerful tool for coordinating and managing multiple AI agents across different terminals and machines, allowing them to message, observe, and react to events in real-time. Key features: - Message each other in real time. - Observe status, commands, and other activity in a terminal. - Control agents by spawning, forking, and killing them. - Connect agents across terminals and machines via MQTT relay. Tech stack: Rust, SQLite database. Why it matters: hcom simplifies the coordination of multiple AI agents, making it easier to manage and coordinate complex AI workflows. Repository: aannoo/hcom Primary language: Rust GitHub stars: 602 GitHub description: Let AI agents message, watch, and spawn each other across terminals.

Overview: MiniSearch is a minimalist web-searching platform that integrates an AI assistant directly into your browser, providing a private and efficient search experience without requiring an API key or third-party access. Key features: - Private by design, no tracking or telemetry. - AI runs entirely in your browser, using WebGPU for GPU-accelerated models and CPU for others. - Supports multiple AI models from curated providers. - Aggregates results from SearXNG, reranks locally, and caches everything within a single Docker container. - Answers can be verified with citations, support for follow-up questions, and can be read aloud. - Local history and analytics, including fuzzy-searchable history, pinning, and session restore. - Fits your workflow, allowing integration with browser defaults, Raycast, and embedding in your own pages. - Secure and customizable with access keys for protection. Tech stack: TypeScript Why it matters: MiniSearch offers a private, efficient, and customizable search experience that integrates an AI assistant directly into your browser, making it ideal for individuals and teams looking for a self-hosted and secure search solution.

Overview: Tomotopy is a Python package that extends the popular `tomoto` library, providing a Gibbs-sampling based topic modeling tool optimized for modern CPUs. It supports various topic models and can significantly speed up training times. Key features: - Supports Latent Dirichlet Allocation (LDA), Labeled LDA (LLDA), Partially Labeled LDA (PLDA), Supervised LDA (SLDA), Dirichlet Multinomial Regression (DMR), Generalized Dirichlet Multinomial Regression (GDMR), Hierarchical Dirichlet Process (HDP), Hierarchical LDA (HLDA), Multi-Grain LDA (MGLDA), Pachinko Allocation (PAM), Hierarchical PA (HPAM), Correlated Topic Model (CTM), Dynamic Topic Model (DTM), Pseudo-document based Topic Model (PTM). Tech stack: C++ Why it matters: Tomotopy is essential for anyone working with topic modeling, as it provides a faster and more efficient alternative to traditional topic modeling libraries like `gensim`. Its SIMD instruction set optimization makes it particularly suitable for modern CPUs, significantly reducing training times and improving overall performance.

Overview: This repository contains open-source speech technology developed by Huawei Noah's Ark Lab, focusing on Grad-TTS, SPIRAL, and DiffVC models. Key features: Grad-TTS, SPIRAL, and DiffVC. Tech stack: Jupyter Notebook. Why it matters: These models are significant contributions to the field of speech technology, offering practical applications in speech synthesis, pre-training, and voice conversion. Repository: huawei-noah/Speech-Backbones GitHub stars: 602

### Overview: This project is an implementation of the BERT model and its related downstream tasks based on the PyTorch framework. It includes detailed explanations of the BERT model and the principles of each underlying task. ### Key Features: - Implementing the BERT model from scratch - Chinese text classification task based on BERT pretraining model - English text implication (MNLI) task based on BERT pretrained model - English Multiple Choice (SWAG) task based on BERT pretrained model - English Question Answering (SQuAD) Task Based on BERT pretrained Model - Training BERT tasks from scratch based on NSL and MLM tasks - Named Entity Recognition task based on BERT pretraining model ### Tech Stack: Python ### Why it Matters: Before using this project, it is recommended to understand the principles of Transformer by studying the examples of Translation, Classification, and Coupling Generation. These examples are provided in the README file.

Overview: This PyTorch implementation is designed for the paper "Matching the Blanks: Distributional Similarity for Relation Learning" published at ACL 2019. It provides models for relation extraction using techniques like ALBERT and BioBERT, focusing on pre-training using the "Matching the Blanks" method. Key features: - ALBERT model for relation extraction - BioBERT model for relation extraction - Pre-training using pre-trained BERT models - Fine-tuning on the SemEval2010 Task 8 dataset Tech stack: Python (3.8+), PyTorch, HuggingFace.co, Spacy Why it matters: This implementation offers a practical tool for relation extraction tasks in natural language processing, leveraging advanced pre-trained models to enhance the accuracy and efficiency of relation extraction in text analysis. Repository: plkmo/BERT-Relation-Extraction GitHub stars: 602 GitHub description: PyTorch implementation for "Matching the Blanks: Distributional Similarity for Relation Learning" paper README excerpt: A PyTorch implementation of the models for the paper ["Matching the Blanks: Distributional Similarity for Relation Learning"](https://arxiv.org/pdf/1906.03158.pdf) published in ACL 2019. Note: This is not an official repo for the paper. Additional models for relation extraction, implemented here based on the paper's methodology. For more conceptual details, please see https://towardsdatascience.com/bert-s-for-relation-extraction

### Overview: This repository provides a PyTorch implementation of Interactive Deep Colorization, a technique for real-time user-guided image colorization. The project is based on the original work by Richard Zhang, Jun-Yan Zhu, and others, and includes training scripts, datasets, and a GUI for interactive colorization. ### Key features: - Real-time user-guided image colorization - PyTorch reimplementation of the original algorithm - Training scripts and pretrained models - Dataset preparation script for ILSVRC 2012 ### Tech stack: Python ### Why it matters: This project offers a practical and efficient way to apply deep learning techniques to real-time image colorization, making it accessible for developers and researchers alike.

Overview: This repository contains a FastAPI skeleton app designed to serve production-ready machine learning models, making it easy and secure to deploy. Key features: - FastAPI Skeleton App - Preconfigured `tox` for rapid expansion - Sample regression model for house price prediction included Tech stack: Python Why it matters: This repository simplifies the process of deploying machine learning models by providing a production-ready FastAPI app, making it ideal for developers looking to quickly integrate ML models into their applications.

### Overview: This repository contains the PyTorch implementation of the DeepLab-ResNet architecture, which was used in the paper "SketchParse: Towards rich descriptions for poorly drawn sketches using multi-task hierarchical deep networks" published at ACM MM 2017. The architecture calculates losses on input images over multiple scales and merges the output score maps from these scales to calculate a total loss. The code includes shared weights for the 3 scales and custom data training capabilities. ### Key features: - PyTorch implementation of DeepLab-ResNet - Calculates losses on input images over 1x, 0.75x, and 0.5x scales - Merges output score maps from 3 scales to calculate total loss - Shared weights for 3 scales in the caffe implementation - Custom data training using `train.py` and `evalpyt.py` - Results almost same as caffe implementation - Reduced memory and time occupied on GPU compared to caffe implementation - Custom data denoted by contiguous label values (0-6 for 7 labels) - Batchnorm parameters fixed during training, `use_global_stats` set to `True` ### Tech stack: Python ### Why it matters: The DeepLab-ResNet architecture is crucial for generating rich descriptions for poorly drawn sketches, enhancing the accuracy and effectiveness of the model's predictions. The shared weights and custom data capabilities make it a valuable tool in the field of sketch-based image understanding.

Overview: EmbedJs is an Open Source Framework designed for personalizing LLM responses, enabling powerful Retrieval-Augmented Generation (RAG) and Large Language Model (LLM) applications in Node.js. Key features: - Segments data into manageable chunks - Generates relevant embeddings - Stores embeddings in a vector database for optimized retrieval - Enables context-aware information extraction and interactive chat conversations Tech stack: TypeScript Why it matters: EmbedJs simplifies the process of integrating LLMs and embeddings into Node.js applications, making it easier to build highly personalized and efficient chatbot and RAG systems.

### Overview: This repository is a curated list of awesome AI tools, frameworks, APIs, software, and resources. It serves as a comprehensive guide for developers looking to explore and utilize cutting-edge AI technologies. ### Key features: - **AI API**: A collection of popular AI APIs including HuggingFace, CivitAI, Mistral AI, Anthropic, Groq, OpenAI, Perplexity, OpenRouter, and Gemini API. - **AI Cloud**: Resources and tools for deploying and managing AI models in the cloud. - **LLMs Backend**: Frameworks and libraries for building AI backend systems. - **LLMs ChatUI**: Tools and resources for creating chat interfaces with AI. - **Stable Diffusion**: High-quality image generation using AI models. - **Code Assistant**: Tools for improving code quality and efficiency. - **Autonomous Agents**: Technologies for creating autonomous agents and robots. - **Knowledge Manager**: Tools for managing and accessing knowledge. - **LLMs Eval**: Evaluation tools and resources for AI models. - **LLMs Framework**: Libraries and frameworks for developing AI models. - **LLMs Training**: Resources and tools for training and optimizing AI models. - **Vector Database**: Tools for storing and accessing large amounts of data. - **AI Course**: Educational resources and courses on AI technologies. - **Audio Gen**: Tools for generating audio content using AI. ### Tech stack: The repository uses a variety of programming languages and frameworks, including Python, JavaScript, and others, to provide a diverse set of tools and resources. ### Why it matters:

Overview: MMMU Benchmark is a massive multi-discipline multimodal understanding and reasoning benchmark designed to evaluate models on college-level subject knowledge and deliberate reasoning tasks. It includes 11,500 meticulously collected questions from college exams, quizzes, and textbooks across six core disciplines and 30 subjects, covering 183 subfields and 32 heterogeneous image types. The benchmark is crucial for evaluating the advanced perception and reasoning capabilities of models, particularly in domains like expert artificial general intelligence (AGI). Key features: - 11,500 meticulously collected multimodal questions from college exams, quizzes, and textbooks across six core disciplines and 30 subjects. - 183 subfields and 32 heterogeneous image types. - Evaluation of 14 open-source LMMs and the proprietary GPT-4V(ision) models. - 599 GitHub stars, indicating significant interest and usage. Tech stack: Python. Why it matters: MMMU Benchmark is essential for evaluating the advanced perception and reasoning capabilities of models, particularly in domains like expert artificial general intelligence (AGI). It provides a rigorous evaluation framework that closely reflects real-world scenarios, offering critical insights for advancing multimodal AI research.

### Overview Timefold Quickstarts provides a comprehensive set of Java and Kotlin implementations for various optimization problems, including vehicle routing, employee scheduling, maintenance scheduling, food packaging, order picking, school timetabling, facility location, conference scheduling, bed allocation scheduling, flight crew scheduling, meeting scheduling, project job scheduling, sports league scheduling, and tournament scheduling. These quickstarts serve as a starting point and inspiration for creating your own applications, with Timefold Solver serving as a library without a UI. ### Key Features - Vehicle Routing - Employee Scheduling - Maintenance Scheduling - Food Packaging - Order Picking - School Timetabling - Facility Location Problem - Conference Scheduling - Bed Allocation Scheduling - Flight Crew Scheduling - Meeting Scheduling - Project Job Scheduling - Sports League Scheduling - Tournament Scheduling ### Tech Stack Java ### Why It Matters These quickstarts are designed to help developers optimize planning and scheduling problems efficiently. They provide a practical way to start using Timefold Solver without the need for a full application, making it easier to understand and implement optimization techniques in various real-world scenarios.

### Overview: FAST, or Fullstack AgentCore Solution Template, is a flexible starter project repository designed for rapid deployment of secure, web-accessible React frontends connected to Amazon Bedrock AgentCore. It simplifies the setup of full-stack applications, enabling developers to focus on the core functionality without the hassle of infrastructure and backend development. ### Key Features: - **React Frontend**: A secure, web-accessible React frontend. - **AgentCore Backend**: A production-ready deployment of any use case on Amazon Bedrock AgentCore. - **Code Interpreter**: Direct integration with Amazon Bedrock AgentCore Code Interpreter for secure Python code execution. - **Gateway Tools**: Lambda-based tools for text analysis and secure Python code execution. - **Deployment**: Out-of-the-box deployment with a minimal setup, including Cdk commands for infrastructure. ### Tech Stack: - Python - AWS Bedrock AgentCore ### Why It Matters: FAST accelerates the development process by handling the undifferentiated heavy lifting of infrastructure setup, allowing delivery scientists and engineers to focus on the core functionality of their applications. The use of AI coding assistants ensures that developers can leverage best practices and code snippets from the documentation, enabling them to build secure and efficient AgentCore applications quickly and easily.