Projects

Overview: This Rust project implements the Jieba Chinese word segmentation algorithm, providing efficient and accurate segmentation of Chinese text. Key features: - Supports default dictionary for quick segmentation. - Enables TF-IDF and TextRank keyword extraction for enhanced analysis. - Compatible with Rust 2015, requiring `extern crate jieba_rs` in your Cargo.toml. Tech stack: Rust. Why it matters: The project offers a fast and accurate solution for Chinese word segmentation, which is essential for natural language processing tasks.

Overview: EAGLE-4 is an inference code designed to adapt to various batch sizes and hardware setups, providing faster language model inference with speculative decoding. Key features: - Speculative decoding for faster inference - Adaptation to different batch sizes and hardware - Speedup of up to 3.4x on Qwen3.8-27B - Measured speedups for EAGLE-4 draft models Tech stack: Python Why it matters: EAGLE-4 offers a flexible solution for high-performance language model inference, particularly beneficial for scenarios with varying batch sizes and hardware configurations. Repository: eagle-hq/EAGLE-4 GitHub stars: 842 GitHub description: Official Inference Code of EAGLE-4 README excerpt: <div align="center"> <img src="docs/assets/Eagle.svg" alt="Eagle" width="180"> <h1>EAGLE-4</h1> <p> <b>One Flexible Drafter for All Batch Sizes and Hardware</b> </p> <p> <a href="EAGLE4_Paper.pdf"><b>Paper (EAGLE-4)</b></a> · <a href="https://huggingface.co/EAGLE-Corp/Qwen3.8-27B-EAGLE4"><b>EAGLE-4 draft for Qwen3.8-27B</b></a> · <a href="docs/README.md"><b>Documentation</b

### Overview: VisRAG 2.0 is an evidence-guided Vision Retrieval-augmented Generation framework that leverages Vision Language Models (VLMs) to enhance the generation of VLMs. It optimizes the training process by applying fine-grained token-level rewards to jointly optimize visual perception and reasoning. ### Key Features: - **Evidence-Guided Retrieval**: Collects per-image evidence by first linguistically observing retrieved images. - **Multi-Image Reasoning**: Reason over the collected evidence to answer multi-image questions. - **End-to-End Pipeline**: A novel VLM-based RAG pipeline that eliminates the parsing process, maximizing the retention and utilization of data information. ### Tech Stack: Python ### Why it Matters: VisRAG 2.0 is a significant advancement in the field of visual language understanding, offering a more efficient and effective way to enhance the generation of VLMs. By directly embedding documents using VLMs and retrieving images to enhance the generation process, it significantly reduces the information loss introduced during the parsing process, leading to improved results and better performance in various applications.

### Overview: This repository is a collection of examples, applications, and tutorials designed to help developers kickstart their GenAI projects. It utilizes LanceDB, a free, open-source, serverless vectorDB, to efficiently store and search documents. The repository integrates seamlessly into the Python ecosystem, making it easy to incorporate vector search capabilities into existing data pipelines. ### Key Features: - **LanceDB**: A free, open-source vectorDB that requires no setup. - **Integration with Python Ecosystem**: Seamless integration into existing data pipelines using libraries like pandas, arrow, and pydantic. - **Native TypeScript SDK**: Provides a serverless function SDK for running vector search in TypeScript. ### Tech Stack: - Jupyter Notebook ### Why It Matters: This repository serves as a comprehensive resource for developers looking to build GenAI applications. It includes examples and tutorials on how to integrate vector search and LLMs into various applications, such as RAG, chatbots, and AI agents, making it easier to create intelligent and efficient AI-driven projects. ### GitHub Stars: 982 ### Repository: lancedb/vectordb-recipes ### Primary Language: Jupyter Notebook ### GitHub Description: Resource, examples & tutorials for multimodal AI, RAG and agents using vector search and LLMs ### Examples: - **Build from Scratch**: Step-by-step guides to create AI applications from scratch. - **Multimodal**: Build apps that process and search across both text and images. - **RAG**: Combine document retrieval with LLM-powered responses

### Overview: Ling Stack is a local-first AI teammate workspace that allows developers to bring their own models and keep project conversations, tool approvals, and results together. It offers a unified interface for managing AI teammates, making it suitable for scenarios involving multiple models or AI teammates. ### Key Features: - **Model and Tool Binding**: Allow teammates to connect their preferred models and tools. - **Conversation and Tool Approval**: Keep all project-related discussions, tool requests, and approvals in one place. - **Project Collaboration**: Facilitate collaborative work by giving teammates model and tool bindings and keeping the outcome with the conversation. ### Tech Stack: TypeScript ### Why It Matters: Ling Stack simplifies the management of multiple AI models and teammates, ensuring all project-related discussions, tool requests, and results are kept in one place, making it easier to collaborate and maintain project coherence.

Overview: QANet is a Tensorflow implementation of Google's QANet, a pre-trained model for machine reading comprehension. It is designed to be efficient and effective for various reading comprehension tasks. Key features: - Efficient implementation of QANet - Reduced hidden size and memory usage compared to the original paper - Improved performance through specific modifications to the original architecture - Pre-trained GloVe embeddings for word representation Tech stack: Python, TensorFlow, NumPy, tqdm, spacy Why it matters: QANet offers a practical and effective solution for reading comprehension tasks, making it a valuable tool for researchers and developers working in this domain. Repository: localminimum/QANet GitHub stars: 985 GitHub description: A Tensorflow implementation of QANet for machine reading comprehension

### Overview: This comprehensive guide is designed to help beginners in programming and machine learning quickly acquire the skills necessary to become an expert in Large Language Models (LLMs). The content is structured to be accessible and engaging, covering everything from foundational concepts to the latest advancements in the field. ### Key features: - **Prerequisites:** If you have 0 programming or AI knowledge, follow the guide on how to get started with machine learning. This will provide a strong foundation. - **YouTube Videos:** Start with short YouTube videos to introduce you to the terminology and concepts essential for LLMs. - **LLM Books and Articles:** For those with a bit of programming experience, explore books and articles to deepen your understanding. - **Online Courses:** Consider taking online courses to further enhance your knowledge and skills. - **Practice:** Practice is key to mastering any skill, so make sure to practice regularly. - **Prompting and Retrieval Augmented Generation (RAG):** Learn how to use prompts and RAG techniques effectively. - **More Resources:** Access a wealth of additional resources including communities, cheat sheets, news, and more. - **Finding a Job in AI/LLM:** Discover tips on how to find a job in the field of AI and LLMs. - **Ethics in AI/LLM:** Learn about the ethical considerations in the development and use of AI and LLMs. - **Additional Resources:** Keep learning and growing with more resources to enhance your knowledge and skills. ### Tech stack: The guide is designed to be accessible and engaging, covering

### Overview: OpenBot is a local-first desktop workspace designed for persistent AI teammates, offering Codex, Claude, and Grok integration through ACP. It includes features like agent creation, persistent workspaces, message queues, file sharing, and browser control, ensuring local-first and secure operations. ### Key Features: - **Persistent Workspaces**: Dedicated workspaces for each agent, ensuring persistent conversations. - **Custom ACP Agents**: Support for custom OpenAI-compatible endpoints and ACP agents. - **Prompt-Driven Agent Creation**: Editable instructions, avatars, and section review before saving. - **FIFO Message Queues**: Persistent and crash-safe message queues with pause, resume, and cancellation options. - **Agent-to-Agent Collaboration**: Messages, replies, reactions, and file transfers between agents. - **Shared Desktop Chats**: Task owners, explicit delegation, shared history, and control over chat channels. - **Embedded Browser**: Accessible through the OpenBot app. - **Security Measures**: Explicit first-launch consent, per-agent context monitoring, and model compaction. - **Optional Accounts**: Supports OpenBot accounts through one-time email codes. ### Tech Stack: TypeScript ### Why it Matters: OpenBot is a development preview, ensuring secure and reliable operations with local-first and offline capabilities, making it suitable for persistent AI teammates. ### Repository: nightly-labs/openbot ### GitHub Stars: 849 ### GitHub Description: A local-first desktop workspace for persistent AI teammates. Run Codex, Claude, and Grok with dedicated workspaces, task queues, file

Overview: TaskStateGuard is a SQLite-backed state reconciliation tool designed for AI agents, backend workers, and local workflows. It reconciles task statuses and delivery states, ensuring tasks are correctly identified as either completed, timed out, or not delivered. Key features: - Reconciles stuck AI-agent tasks after restarts and timeouts. - Preview SQLite changes, close stale delivery states, and never guess success. - Solves task state reconciliation and stuck task recovery issues. - Three tools for different use cases: TaskStateGuard, ChatArchiveGuard, and ArtifactProof. Tech stack: Python Why it matters: TaskStateGuard provides a reliable way to track and manage task statuses, ensuring accurate and auditable results, making it essential for maintaining the integrity of AI workflows and ensuring data accuracy in various applications. Repository: MaxHu-xuan/task-state-guard Primary language: Python GitHub stars: 957

Overview: This project presents a novel method called Geometric Transformer for fast and robust point cloud registration, which addresses the challenge of extracting accurate correspondences in point cloud registration. The method uses geometric feature learning to match superpoints and is designed to be robust against low-overlap scenarios and rigid transformations. Key features: - Geometric Transformer for robust superpoint matching - High matching accuracy without the need for RANSAC - 100x acceleration compared to traditional methods - Improved inlier ratio and registration recall on the 3DLoMatch benchmark Tech stack: Python Why it matters: The Geometric Transformer method offers a significant improvement in point cloud registration accuracy, making it highly relevant for applications requiring high precision in 3D data processing, such as autonomous driving, robotics, and augmented reality. Repository: qinzheng93/GeoTransformer GitHub stars: 984

Overview: This repository contains resources on the applications of multi-modal learning in medical imaging, focusing on papers related to large language models (LLM). Key features: - Papers on the integration of LLMs in medical imaging - Papers on the evaluation of medical vision-language models (Med-VLMs) - Papers on the optimization of multi-agent collaboration for multimodal medical reasoning Tech stack: Python, Markdown, and GitHub Why it matters: This repository serves as a comprehensive resource for researchers and practitioners interested in the latest advancements in the intersection of multi-modal learning and medical imaging, with a particular focus on the use of large language models (LLM) in medical applications. Repository: richard-peng-xia/awesome-multimodal-in-medical-imaging GitHub stars: 984

### Overview: Git designed for AI agents. ### Key features: - **Always-online clients**: Fetch exactly what you need. - **Source control as an API**: History runs on the server. - **Landing without a merge queue**: Rebase a change onto the current tip and land it with a compare-and-swap. - **Automatic conflict resolution**: Resolved by a model you choose, or a deterministic tool, or none. - **Changeset log**: Every accepted reference update is recorded in a dense, ordered log. ### Tech stack: Go ### Why it matters: Git designed for AI agents.

Overview: K-BERT is a language representation model that leverages knowledge graphs to enhance language understanding and classification tasks. Key features: - Integrates knowledge graph information for improved classification and NER performance. - Supports text classification and named entity recognition (NER) tasks. - Utilizes the UER framework for efficient model training and inference. Tech stack: Python, Pytorch, argparse, UER framework. Why it matters: K-BERT enhances language representation by incorporating knowledge graph information, making it particularly useful for applications requiring accurate classification and entity recognition tasks, especially in domains where context and knowledge are crucial.

Overview: A lightweight, zero-dependency CLI tool designed to manage and activate capabilities for AI coding assistants like Claude Code, Cursor, and Trae. Key features: - Skills (modular prompting) - Agents (expert workflow roles) - Commands (slash commands) Tech stack: TypeScript Why it matters: Simplifies the management and activation of AI coding assistant capabilities, making it easier to integrate and use these tools in projects or global system configurations. Repository: guanyang/open-agent-hub GitHub stars: 979 GitHub description: A lightweight, zero-dependency CLI tool to manage and activate capabilities for AI coding assistants (such as Claude Code, Cursor, Trae, etc.). Tech stack: TypeScript Why it matters: Simplifies the management and activation of AI coding assistant capabilities, making it easier to integrate and use these tools in projects or global system configurations.

### Overview: Doop is an open-source alternative to Paper.design, a multiplayer design canvas where humans and AI agents collaborate in real-time. It allows for live design, streaming of designs, and the ability to connect with various MCP clients, including Claude Code, ChatGPT, and OpenCode. ### Key Features: - **Design with Agents, Not Prompts-and-Refresh**: Connect Claude Code or any MCP client once, then watch it sketch, stream, and self-review designs on your canvas. - **Built-in Doop Agent**: Queue a card or mention a role to have the agent design on its own, using the user's connection key (ChatGPT or OpenAI). - **Pages within a Canvas**: Organize frames into named pages, edit frames, and manage pages with the Layers panel. Supports Undo/Redo, and can move frames between pages. - **True Multiplayer**: Live cursors, presence, per-frame editing indicators, undo/redo, comments pinned to elements, and an activity feed. - **Design Memory**: Pin exemplar frames, capture decisions, and let the distiller propose durable style rules. - **Private by Default**: Invite collaborators by email or share the link per canvas; agents inherit their human's access. - **Self-host in One Command**: Use `docker compose up` or `bun run dev` for zero configuration, including embedded Postgres and no external services required. ### Tech Stack: TypeScript ### Why It Matters: Doop is a powerful tool for designers and developers looking to collaborate with AI agents in real

### Overview: The Plato Research Dialogue System is a comprehensive framework designed for the development of conversational AI agents. It supports interactions through speech, text, or dialogue acts and allows agents to interact with data, human users, or other agents in a multi-agent setting. The system is highly flexible and can be trained offline or online, making it suitable for various environments. ### Key Features: - **Domain**: Defines the context and goals of the conversation. - **Controller**: Manages the flow of the conversation and interacts with the agent. - **Agent**: Processes input, generates output, and interacts with the environment. - **Rule-based modules**: Handle simple, predefined rules. - **Trained modules**: Use existing models for more complex interactions. ### Tech Stack: - Python ### Why it Matters: Plato provides a robust platform for developing and training conversational AI agents, making it ideal for researchers and developers looking to create intelligent dialogue systems. Its flexibility and ease of use make it accessible to a wide range of users, from beginners to advanced practitioners. ### Repository: - **Repository**: `uber-archive/plato-research-dialogue-system` - **GitHub Stars**: 980 - **GitHub Description**: "This is the Plato Research Dialogue System, a flexible platform for developing conversational AI agents." ### Installation: - **Installing Plato from source code (Recommended)**: ```bash pip install git+https://github.com/uber-archive/plato-research-dialogue-system.git ``` - **Common Issues During Installation**: - Ensure Python is

### Overview: This curated list of 3D Morphable Model (3DMM) software and data aims to compile shared resources and algorithms related to 3DMMs. The list serves as a repository for researchers and enthusiasts interested in the field, encouraging contributions through pull requests. ### Key Features: - **Software and Data Repositories**: Includes software and data repositories for various 3DMM models, such as Google Group, Face Models, Body Models, and Other Models. - **Registration and Construction Tools**: Provides pipelines for large-scale facial models, face registration, shape-aware surface reconstruction, and more. - **Model Construction and Alignment**: Offers tools for constructing and aligning 3DMM models from diverse raw scan data. ### Tech Stack: - **Python**: Used for implementing and managing the software repositories. - **MATLAB**: Utilized for some of the registration and alignment tools. ### Why It Matters: This list serves as a valuable resource for researchers and practitioners in the field of 3DMMs, providing a centralized platform for sharing and accessing the latest techniques and models. It also encourages collaboration and innovation through the contributions of the community, making it an essential resource for anyone interested in advancing the field of 3DMMs.

### Overview: Claude Skills is a comprehensive set of 385 AI skills designed to enhance the capabilities of any team member, from engineers to C-level executives. It includes 76 expert agents and 906 Python tools, making it a versatile toolkit for every project. ### Key Features: - **385 Skills**: A wide range of skills tailored for different business domains. - **76 Expert Agents**: Advanced tools for specific tasks, such as compliance audits and board preparation. - **906 Python Tools**: Essential tools for developers and engineers to streamline their work. ### Tech Stack: - **HTML**: The primary language used for creating the project. ### Why It Matters: Claude Skills provides a powerful set of tools that can significantly enhance the productivity and effectiveness of teams across various industries. Whether you need to prepare for a board meeting, perform a compliance audit, or just need a quick way to generate a metrics dashboard, Claude Skills has the solution. ### Repository: - **Repository**: `borghei/Claude-Skills` - **GitHub Stars**: 965 - **GitHub Description**: 385 AI skills, 77 expert agents, and 900 stdlib Python tools for every team: engineering, PM, marketing, C-level, compliance, business ops, research, and a LinkedIn toolkit. Installs on Claude Code, Cursor, Codex, Gemini, Copilot, and 6 other AI assistants. ### Add It to Your Assistant: - **Claude Desktop App**: Open **Customize** (bottom

Overview: Otto is an intuitive machine learning assistant designed for beginners, helping them quickly implement machine learning projects with minimal domain knowledge. Key features: - Task inference: Otto narrows down the high-level task from a simple objective statement. - Machine learning tools: Supports models for regression, classification, and natural language processing. - Educational experience: Users are guided through each stage of the process with Otto explaining terminology. - Sample datasets: Provides sample datasets for regression and natural language processing. - Custom dataset option: Experience Otto-matic model recommendation and an interactive neural network designer. Tech stack: JavaScript Why it matters: Otto simplifies the process of machine learning for beginners, making it accessible and easy to use.

Overview: YouTokenToMe is an unsupervised text tokenizer designed for high computational efficiency, focusing on fast Byte Pair Encoding (BPE) implementation. It is significantly faster than popular libraries like Hugging Face, fastBPE, and SentencePiece, achieving up to 60 times faster performance in training and tokenization. Key features: - Multithreading for efficient training and tokenization - O(N) complexity algorithm, where N is the length of the training data - Highly efficient C++ implementation - Python wrapper and command-line interface - BPE-dropout feature from Provilkov et al., 2019 Tech stack: C++ Why it matters: YouTokenToMe is optimized for computational efficiency, making it suitable for applications requiring fast text processing, such as natural language processing, text generation, and other tasks where speed is crucial. Its superior performance makes it a valuable tool for developers looking to enhance the speed and efficiency of their text processing pipelines.