Projects

### Overview: DBpedia Spotlight is a tool for automatically annotating mentions of DBpedia resources in text. ### Key features: - Automatically annotates mentions of DBpedia resources in text. - Links mentions to their global unique identifiers in DBpedia. - Supports multiple languages and multilingual text processing. ### Tech stack: Scala ### Why it matters: - Provides efficient and accurate entity extraction for multilingual text. - Facilitates linking of mentions to global identifiers in DBpedia. - Enables researchers and developers to easily integrate DBpedia Spotlight into their applications.

Overview: COMPASS Skills is a comprehensive system designed for AI agents, providing ten local skills tailored for various tasks and collaboration needs. These skills are essential for personalizing interactions and managing workflows efficiently. Key features: - Task Clarifier: Aligns goals, scope, evidence, acceptance criteria, and risk boundaries. - Task Forest: Maintains a task forest with goals, subtasks, dependencies, progress, deviations, and conversation history. - Pause and Resume: Cooperatively pauses unfinished work at a safe boundary and resumes it from a precise checkpoint. - Session Handoff Prompt: Compresses the current AI conversation's goal, progress, constraints, and next steps into a paste-ready prompt. - User Profile Keeper: Maintains a local, auditable, correctable collaboration profile. - Run History Skill Builder: Turns completed or repeatedly refined run history into a new skill package. - Run History Skill Upgrader: Automatically turns session evidence into an upgrade plan, preserving changes only after explicit approval. - Clear Communication: Organizes explanations, reports, comparisons, and instructions around the reader's task. - Academic Humanizer: Helps write or revise academic prose by removing formulaic patterns and restoring a natural scholarly voice. - Assess Interview Candidate: Turns a resume and job description into an auditable evidence layer and a concise three-part offline interviewer report. Tech stack: Python Why it matters: COMPASS Skills provides a set of ten skills that enable AI agents to manage and collaborate more effectively, ensuring tasks are aligned with goals and constraints, and interactions are clear and understandable.

Overview: Scrawl is a powerful tool that allows developers to visually communicate their ideas by recording and annotating screen interactions, voice commands, and marks. Key features: - Record screen, voice, and marks - Generate a compact bundle with contact-sheet, transcript, and interaction timeline - Support for various agents and agents skills - Ability to open, inspect, and export bundles Tech stack: Rust Why it matters: Scrawl simplifies the process of documenting and sharing complex interactions, making it easier for developers to communicate their ideas and collaborate effectively. Repository: kitlangton/scrawl GitHub stars: 726 GitHub description: Show an agent a problem: mark up, record, send

### Overview: 本项目是程序员鱼皮精心设计的AI编程学习与求职辅导机器人，基于Spring Boot 3.5.3、Java 21、LangChain4j和阿里云通义千问构建。它涵盖了AI大模型接入、LangChain4j核心特性、流式对话、Prompt工程、RAG检索增强、向量数据库、工具调用、MCP模型上下文协议、Web爬虫、安全防护、Vue.js前端开发、SSE服务端推送等企业级AI应用开发技术。通过实战项目，帮助开发者掌握AI时代必备技能，熟悉LangChain框架，提升编程学习效率和求职竞争力，成为企业需要的AI全栈开发人才。 ### Key features: - **LangChain4j集成**: 采用业界领先的AI应用开发框架。 - **通义千问模型**: 基于阿里云大模型，专业可靠。 - **流式响应**: 实时打字机效果，提升用户体验。 - **RAG检索增强**: 结合本地知识库，提供精准答案。 - **MCP协议支持**: 模型上下文协议，增强AI能力。 - **面试题搜索**: 实时抓取最新面试题目。 - **Web爬虫工具**: 获取实时技术资讯。 ### Tech stack: - **Java**: JDK 21+ - **Node.js**: 16.0+ - **Maven**: 3.6+ - **通义千问API**: 需申请API密钥 - **

### Overview: MONAI Generative Models is a repository designed to facilitate the prototyping and integration of generative models into the MONAI framework. It provides a comprehensive set of tools for training, evaluating, and deploying generative models, including various network architectures, loss functions, and metrics. ### Key features: - Network architectures: Diffusion Model, Autoencoder-KL, VQ-VAE, Autoregressive transformers, and (Multi-scale) Patch-GAN discriminator. - Diffusion Model Noise Schedulers: DDPM, DDIM, and PNDM. - Losses: Adversarial losses, Spectral losses, and Perceptual losses. - Metrics: Multi-Scale Structural Similarity Index Measure (MS-SSIM) and Fréchet inception distance (FID). - Diffusion Models, Latent Diffusion Models, and VQ-VAE + Transformer Inferers classes. - MONAI-compatible trainer engine (based on Ignite) for training models. - Tutorials covering various applications such as VQ-VAEs, VQ-GANs, VQ-VAE + Transformers, AutoencoderKLs, Diffusion Models, and Latent Diffusion Models. - Train diffusion model for conditional image generation with classifier-free guidance. - Comparison of different diffusion model schedulers. - Anomaly Detection using VQ-VAE + Transformers and Diffusion Models. - Inpainting with diffusion model using Repaint method. - Super-resolution with Latent Diffusion Models using Noise Conditioning Augmentation. ### Tech stack: Python 3.

### Overview: UniK3D: Universal Camera Monocular 3D Estimation is a project that aims to estimate 3D points from a single monocular image using a universal camera model. The project was accepted at CVPR 2025 and has been recognized for its innovative approach to 3D reconstruction from monocular images. ### Key Features: 1. **Universal Camera Model**: The project utilizes a universal camera model to estimate 3D points from a single monocular image, making it applicable to a wide range of scenarios. 2. **Model Release**: The project includes models released for training and inference, ensuring reproducibility and ease of use. 3. **Demo and Visualization**: A Gradio demo and Hugging Face demo are available, allowing users to test the project without the need for installation. 4. **Training and Inference Code**: The project provides training and inference code, making it easy to experiment with different models and configurations. ### Tech Stack: - **Python**: The primary programming language used for implementing the project. - **CUDA**: The project leverages CUDA for efficient computation, particularly for the inference process. ### Why It Matters: UniK3D offers a novel solution for 3D reconstruction from monocular images, making it a valuable contribution to the field of computer vision. Its universal camera model and efficient inference process make it suitable for real-world applications and research projects.

Overview: This repository contains code exercises and materials for Udacity's Computer Vision Nanodegree program, focusing on various computer vision applications and techniques. The exercises depend on specific software packages, and it is recommended to create a local environment with these dependencies. Key features: - Code exercises and materials for Udacity's Computer Vision Nanodegree program - Dependencies on specific software packages for running the exercises - Tutorial notebooks demonstrating or challenging computer vision applications and techniques Tech stack: Jupyter Notebook Why it matters: The repository provides practical exercises and materials for learning computer vision, allowing students to apply theoretical knowledge to real-world problems.

### Overview: This repository contains the PyTorch implementation of the CVPR 2018 paper "Unsupervised Feature Learning via Non-parametric Instance Discrimination". The code provides an updated instance discrimination model with memory bank and NCE loss, achieving state-of-the-art performance on ImageNet. ### Key Features: - **Unsupervised Learning**: Formulated from a non-parametric perspective. - **Compact Feature Encoding**: Achieves feature encodings as compact as 128 dimensions. - **Benefit of Supervised Techniques**: Leverages advanced architectures and techniques from supervised learning. - **Nearest Neighbor Classifier**: Runs seamlessly with nearest neighbor classifiers. ### Tech Stack: - Python ### Why It Matters: Unsupervised learning is a powerful technique that can extract meaningful features from unlabeled data, offering a significant advantage over traditional supervised learning methods. The implementation provided in this repository showcases the effectiveness of unsupervised learning in achieving high accuracy with minimal labeled data. ### GitHub Stars: 759 ### Contact: For any questions, please contact: - **Zhirong Wu**: xavibrowu@gmail.com

### Overview: Chaplin is a real-time silent speech recognition tool that reads your lips in real-time and types what you silently mouth. It relies on a model trained on the Lip Reading Sentences 3 dataset as part of the Auto-AVSR project. ### Key features: - **Real-time Lip Reading**: Captures your lips in real-time and types what you silently mouth. - **Local Running**: Runs fully locally without requiring any external dependencies. - **Model Training**: Utilizes a model trained on the Lip Reading Sentences 3 dataset for high accuracy. ### Tech stack: Python ### Why it matters: Chaplin is a valuable tool for those who want to improve their lip-reading skills or need a reliable way to capture and translate silent speech. Its local running capability makes it accessible on any device with a camera and internet connection. Repository: amanvirparhar/chaplin Primary language: Python GitHub stars: 759

### Overview: MASCOT is a multi-agent framework designed to enhance social interaction and well-being, addressing the needs of individuals who lack social contact. It balances individual persona consistency with global discourse dynamics to create socio-collaborative companions. ### Key Features: - 🎭 Persona Fidelity: Fine-tunes agents for strict, unique personas, preventing persona collapse. - 🤝 Interaction Synergy: Guides turn-taking for complementary dialogue, avoiding social sycophancy and echo chambers. - 📦 Self-contained pipeline: Episode generation, rubric judging, preference-data construction, reward modeling, and training all integrated. - 🪶 Low-resource bi-level optimization: Speaker and Director policies train with LoRA adapters, reward-model checkpoints, and explicitly merged checkpoints. ### Tech Stack: Python ### Why it Matters: MASCOT targets user-agent interaction quality in affective and collaborative settings, providing emotional support and workplace meetings. It outperforms prior multi-agent systems by improving persona consistency and social contribution, achieving a 69% human preference score across empathetic dialogues and QMSum metrics. ### Repository: hello-diana/MASCOT ### GitHub Stars: 741 ### GitHub Description: EMNLP 2026 Main Conference Paper: MASCOT: Towards Multi-Agent Socio-Collaborative Companion Systems (https://arxiv.org/abs/2601.14230)

### Instruction following - [NeurIPS'23] [Natural Language-conditioned Reinforcement Learning with Inside-out Task Language Development and Translation](https://arxiv.org/abs/2302.09368) ### Build agent based on World model - [NeurIPS'23] [Guide Your Agent with Adaptive Multimodal Rewards](https://openreview.net/attachment?id=G8nal7MpIQ&name=pdf) [[project](https://sites.google.com/view/2023arp)] ### Language as knowledge - [NeurIPS'23] [Natural Language-conditioned Reinforcement Learning with Inside-out Task Language Development and Translation](https://arxiv.org/abs/2302.09368) - [NeurIPS'23] [Guide Your Agent with Adaptive Multimodal Rewards](https://openreview.net/attachment?id=G8nal7MpIQ&name=pdf) [[project](https://sites.google.com/view/2023arp)] ### LLM as a tool - [NeurIPS'23] [Natural Language-conditioned Reinforcement Learning with Inside-out Task Language Development and Translation](https://arxiv.org/abs/2302.09368) - [NeurIPS'23] [Guide Your Agent with Adaptive Multimodal Rewards](https://openreview.net/attachment?id=G8nal7MpIQ&name=pdf) [[project](https://sites.google.com/view/20

Overview: GRID is a state-of-the-art framework for generative recommendation systems that leverages semantic IDs to improve recommendation quality. It uses LLMs to generate item embeddings and transforms them into hierarchical semantic IDs, which are then used to generate recommendation sequences through transformer-based models. Key features: - Embedding Generation with LLMs - Semantic ID Learning - Generative Recommendations Tech stack: Python Why it matters: GRID addresses the limitations of traditional recommendation systems by incorporating semantic IDs, leading to more accurate and personalized recommendations. This approach enhances user engagement and satisfaction by providing contextually relevant and personalized suggestions. Repository: snap-research/GRID GitHub stars: 758 GitHub description: GRID: Generative Recommendation with Semantic IDs

### Overview: This is an unofficial implementation of the 3D Gaussian Splatting algorithm for real-time radiance field rendering, developed using the Taichi language. Taichi is designed for high-performance computing and Python, offering a balance between productivity and performance. ### Key Features: - **Training**: The algorithm processes multiple views of images, a sparse point cloud, and camera poses to train the point cloud and output a dense point cloud with additional features like covariance and color information. - **Inference**: It takes the dense point cloud with extra features and any camera pose to render images from the new pose. ### Tech Stack: - Taichi, a language for high-performance computing. ### Why It Matters: - **Performance**: Taichi achieves comparable or better performance in Python compared to CUDA implementations, making it easier to read and maintain. - **Flexibility**: Taichi supports multiple backends, including CUDA, OpenGL, Metal, and more, allowing for flexibility in future development. ### Current Status: - The repo has been tested with the provided dataset and achieves slightly higher PSNR than the official implementation, with fewer points. - The training/inference speed is slower than the official implementation, but the results are still competitive. ### Additional Example: - The repo has been tested on the truck dataset, achieving higher PSNR and SSIM values compared to the official implementation. ### Dataset Results: | Dataset | Source | PSNR | SSIM | #points | | --- | --- | --- | --- | --- | | Truck(7k) | Paper | 23.51 |

Overview: This repository provides a Unity3d binding for OpenAI's Whisper automatic speech recognition model, allowing real-time inference on local machines. It supports multilingual support, translation capabilities, and runs without internet connection. Key features: - Multilingual support for around 60 languages - Translation between languages - Different model sizes for speed and accuracy trade-offs - Runs on local devices without internet connection - Free and open-source, suitable for commercial projects Tech stack: Unity3d Why it matters: This tool simplifies the integration of Whisper into Unity projects, making it accessible for developers and allowing for real-time speech-to-text functionality on any device. Repository: Macoron/whisper.unity GitHub stars: 756

Overview: FaceSwap is an app that allows users to swap their face with another person's face captured in an image, using Python for its implementation. Key features: - Face swapping functionality - Python-based implementation - Uses face alignment, Gauss Newton optimization, and image blending for face swapping Tech stack: Python Why it matters: FaceSwap provides a practical application of computer vision techniques, demonstrating how to manipulate images and align facial features, which is valuable for both educational purposes and real-world applications in areas such as digital media and entertainment.

### Overview: This repository contains a collection of 16,809 open-source tools that were reverse-engineered from closed apps, APIs, devices, protocols, file formats, games, and AI products. The tools are maintained and can be used by anyone to enhance their skills or contribute to the community. ### Key Features: - Search site + agent skill + MCP server, refreshed nightly - 16,809 projects with 11.2M GitHub stars - Top languages: Python (3,792), C++ (2,610), C (2,495), JavaScript (983), C# (902) - Categories include AI Tools & LLM Clients, Game Decompilations & Reimplementations, Devices, IoT & Hardware, Unofficial Web/App API Clients, and Messaging & Social Apps ### Tech Stack: Python ### Why It Matters: This repository is a valuable resource for developers looking to enhance their skills in reverse engineering, open-source development, and AI tooling. It serves as a comprehensive index for anyone interested in exploring and contributing to the open-source ecosystem.

Overview: Anton is a self-improving agent designed to handle a variety of tasks, including sending emails, managing calendars, automating reporting, and more. It is integrated into MindsHub Cowork, a unified workspace, and can also be run standalone. Key features: - Sends emails - Manages calendars - Automates reporting - Automates workflows - Researches and synthesizes data - Prototypes data pipelines - Performs system administration tasks Tech stack: Python Why it matters: Anton is a powerful tool that can automate a wide range of tasks, making it easier for users to focus on their core responsibilities while still achieving their goals. Repository: mindsdb/anton Primary language: Python GitHub stars: 757 GitHub description: self-improving agent

### Overview: This is a curated library of 34 specialized Claude agent configurations organized into 7 professional categories. Each agent is a detailed markdown file containing purpose, core responsibilities, key skills, communication style, example prompts, and related agents. ### Key Features: - **Agent Selection Guide**: Find agents by role, task, or team size. - **Model Selection Matrix**: 30+ task-specific recommendations with costs. - **Cost Optimization**: Save up to 82% on Claude API costs. - **Quick Agent Finder**: Guides for finding agents by role or common task. - **Customization**: Agents can be easily copied, pasted, and customized for any professional role. ### Tech Stack: Python and Claude API. ### Why it Matters: This library transforms Claude into a domain expert with consistent behavior and task-specific knowledge, making it easier and faster to write prompts and achieve consistent quality, faster results, and expertise on demand.

### Overview: Codanna is a local code intelligence and semantic code search MCP server for AI coding agents. It indexes your repository on disk and serves symbol search, semantic search, call graphs, dependency tracking, document RAG, and impact analysis to various MCP-compatible clients. ### Key features: - **Symbol Search**: Returns the identity, signature, and documentation of symbols in your code. - **Call Graph**: Provides a call graph of symbols in your code. - **Dependency Tracking**: Tracks dependencies of symbols in your code. - **Document RAG**: Retrieves relevant documents based on the query. - **Impact Analysis**: Provides impact analysis for symbols in your code. - **Persistent MCP Server**: Supports session-long agent workflows. - **One-Shot CLI**: Supports instant answers when LSP is too slow. ### Tech stack: Rust ### Why it matters: Codanna offers a powerful and efficient way to enhance AI coding agents by providing comprehensive code intelligence and analysis capabilities directly within your development environment.

### Overview: agentacct is a powerful tool that allows developers to track the activities and costs of their coding agents. It provides a detailed breakdown of each task, including tools used, files changed, tests run, time, and tokens spent. The local-first dashboard offers a seamless user experience without requiring login or leaving the user's computer. ### Key Features: - **Detailed Activity Tracking:** Tracks the activities and costs of coding agents across projects and clients. - **Work Receipts:** Records the steps taken, tool calls, and estimated costs for each task. - **Timeline View:** Displays the activity timeline from attempt to result, showing steps and check details. - **No Login or Telemetry:** Ensures privacy and security by not requiring any login or leaving the user's computer. ### Tech Stack: - Python ### Why It Matters: agentacct provides a clear and comprehensive view of what coding agents are doing, helping developers understand their work processes and costs. It simplifies tracking and management, making it easier to manage and optimize agent usage without the need for complex setup or external dependencies. ### Repository: - **Repository:** `mikehasa/agentacct` - **GitHub Stars:** 756 - **Primary Language:** Python - **GitHub Description:** See what your coding agents did and what it cost. Breaks each task down into work steps — tools used, files changed, tests run, time and tokens spent. Local-first dashboard for Claude Code, Codex, OpenCode, and more. No login, no telemetry. ### Install: - **Install via Pip