Projects

Overview: SearChat is an AI-powered conversational search engine built with TypeScript, supporting multiple search engines and deep research functionality. It integrates with OpenAI, Anthropic, Gemini, and SearXNG, offering a modern interface experience with responsive design, dark/light themes, and internationalization support. Key features: - Multi-model support for OpenAI, Anthropic, Gemini APIs - Multiple search engines integration - Conversational search with chat history - Deep research mode with structured reports - Support for SearXNG, Bing, Google, Tavily, Exa, Bocha, and ChatGLM Web Search Tech stack: TypeScript, Node.js + Koa backend, Vue 3 + TypeScript frontend Why it matters: SearChat is a versatile tool for anyone looking to enhance their search capabilities with AI-powered insights and a modern user interface.

### Overview: This repository contains a curated list of academic papers focused on generative Information Extraction (IE) using Large Language Models (LLMs). The papers are organized into various tasks and techniques, providing a comprehensive overview of the field. ### Key features: - **Information Extraction Tasks:** - **Named Entity Recognition (NER):** - **Entity Typing:** Models that recognize and categorize specific entities. - **Entity Identification & Typing:** Models that identify and categorize entities. - **Information Extraction Techniques:** - **Supervised Fine-tuning:** Fine-tuning LLMs on labeled datasets. - **Few-shot:** Using a small number of labeled examples to train models. - **Zero-shot:** Models that can recognize entities without explicit training on the target domain. - **Data Augmentation:** Techniques to enhance the diversity of training data. - **Prompts Design:** Crafting specific prompts to guide LLMs towards the desired output. - **Constrained Decoding Generation:** Limiting the output to specific formats or structures. ### Tech stack: - Python, Jupyter Notebook, and various LLMs (e.g., GPT-3, Qwen, etc.). ### Why it matters: This repository serves as a valuable resource for researchers and practitioners interested in the latest advancements in generative Information Extraction using LLMs. It provides a structured way to explore and stay updated with the most relevant papers in the field. Repository: quqxui/Awesome-LLM4IE-Papers GitHub stars:

Overview: This guide aims to keep developers updated with the latest news, papers, discussions, and tutorials in the rapidly evolving fields of AI, Machine Learning, Deep Learning, and Computer Vision. Key features: - Stay connected to local communities through forums and meetups. - Access to essential resources and updates from various countries. Tech stack: None specified. Why it matters: This guide is essential for developers looking to stay current with the latest trends and advancements in AI, Machine Learning, Deep Learning, and Computer Vision.

Overview: Experts.js simplifies the creation and deployment of OpenAI's Assistants and Tools, making it easier for developers to create advanced Multi AI Agent Systems with expanded memory and attention. Key features: - Simplifies managing Run objects - Introduces Assistants as Tools - Supports instructions up to 256,000 characters - Integrates with 128 tools - Utilizes Vector Store API for efficient file search Tech stack: JavaScript Why it matters: Experts.js streamlines the use of OpenAI's Assistants API, making it more accessible and easier to integrate into existing projects, thus enabling developers to create more complex and advanced AI systems.

Overview: This repository contains a catalog of API tools that an agent can call, including 660+ generative-media models hosted by muapi and a growing set of third-party tools. The catalog is designed to provide detailed information on pricing, cost, and usage without requiring the user to sign up or install anything. Key features: - `models/` entries: One of muapi's own hosted generative-media models. - `providers/` entries: A third-party API a contributor already uses. - Editable by PR: This is the open contribution path. Tech stack: Python Why it matters: The tools worth calling from an agent are scattered across dozens of vendors, each with its own docs, auth quirks, and pricing page. This catalog puts the facts that matter—auth shape, real pricing, a captured example response—into one consistent shape, making it easier for agents to know exactly what a tool needs before opening its docs. Repository: Anil-matcha/awesome-agent-apis GitHub stars: 1057

### Overview: VisCPM 是一个基于 CPM 基础模型的开源多模态大模型系列，支持中英双语的多模态对话能力和文到图生成能力。该系列模型在中文多模态开源模型中达到最佳水平，通过融合视觉编码器和视觉解码器实现高效的视觉-语言表示对齐。 ### Key features: - **多模态对话模型 VisCPM-Chat**: 支持面向图像的中英双语多模态对话。 - **文生图模型 VisCPM-Paint**: 可生成高质量的中文和英文文本到图像的转换结果。 - **预训练和指令精调**: 模型经过预训练和指令精调两阶段训练，以支持大规模视觉-语言表示的高效对齐。 ### Tech stack: Python ### Why it matters: 得益于语言模型基座 CPM-Bee 的优秀双语能力，VisCPM 在中英双语的多模态对话和文到图生成中取得了亮眼的效果，推动了多模态大模型开源社区和相关研究的发展。

Overview: This demo implements offline speech recognition and speaker identification for Android applications using the Vosk library. Key features: - Offline speech recognition - Speaker identification Tech stack: Java Why it matters: This demo provides a practical solution for developers looking to integrate speech recognition capabilities into their Android applications without the need for a real-time internet connection.

### Overview: Algebrica is a free, ad-free, university-level mathematics knowledge base. This repository hosts the source content of the entries published on the site, released progressively in Markdown format and reusable under a non-commercial license. ### Key Features: - The original source files of the entries written for Algebrica in Markdown and LaTeX format. - Fully editable SVG illustrations, including both vector structures and text elements, which can be modified directly through the source code. - A semantic JSON layer containing definitions, properties, lemmas, theorems, and other formal mathematical structures associated with the entries, designed to be portable, reusable, and openly accessible. - Markdown-based flowcharts describing the main logical and procedural steps behind common mathematical processes and problem-solving methods. ### Tech Stack: Markdown and LaTeX ### Why it Matters: Algebrica is built with the principles of clarity, structure, and conceptual coherence, making it an essential resource for students and educators alike. The progressive nature of the project ensures that the content remains up-to-date and relevant, while the open-source approach allows for continuous improvement and expansion.

Overview: This repository is a curated collection of free and open-source resources for learning and building AI agents, including autonomous agents, LLM-powered agents, multi-agent systems, and agentic AI frameworks. Perfect for beginners and advanced developers seeking comprehensive learning and practical examples. Key features: - Comprehensive curated lists and awesome collections of AI agents, frameworks, projects, and tools. - Structured course for beginners covering fundamentals, architecture, tool use, memory, and planning. - Hands-on learning with 45+ Jupyter notebook implementations, including conversational AI, multi-agent systems, and more. - Structured video tutorials for practical automation and no-code/low-code agent tutorials. Tech stack: Python, Jupyter notebooks, LangChain, LangGraph, CrewAI, AutoGen, OpenAI Swarm, Discord, Reddit. Why it matters: This repository serves as a valuable resource for anyone looking to learn and build AI agents, providing a wide range of resources and practical examples to enhance their understanding and skills in the field.

### Overview: Composing and running AI agents with Docker Compose, designed to simplify the deployment and orchestration of various AI models and tools. This project provides a collection of examples that showcase how to use Docker Compose to manage and run open-source LLMs, tools, and agent runtimes. ### Key Features: - Multi-Agent Fact Checker - Agent Summarizing GitHub Issues - Chat-UI for Mixing MCPs and Models - Marketing Strategy Agent - Multi-Agent Fact Checker - Multi-Agent Fact Checker - SQL Agent - Travel Agent - Brave Search - Sock Store Agent ### Tech Stack: TypeScript ### Why It Matters: This project streamlines the process of deploying AI agents, making it easier for developers to integrate and run their models without the complexities of setting up environments manually. The examples provided cover a wide range of use cases, from multi-agent systems to single-agent models, ensuring flexibility and practical application.

### Overview: Parallel Code allows developers to run multiple AI coding tools (Claude Code, Codex, and Gemini) side by side, each in its own git worktree. This enables parallel execution, isolated changes, and easy review of diffs. ### Key Features: - **Multiple Agents**: Run up to ten AI coding tools in parallel. - **Ten Branches**: Each agent operates in its own git branch. - **One afternoon**: Dispatch multiple agents to complete tasks simultaneously. ### Tech Stack: - **TypeScript**: The primary programming language used. - **Electron**: For the user interface and cross-platform compatibility. ### Why It Matters: - **Free and Open Source**: No extra subscription required, MIT licensed. - **Isolated Changes**: Keeps every change isolated and reviewable. - **Parallel Execution**: Runs agents in parallel, avoiding conflicts. - **Easy Management**: Switch context without losing momentum. - **Keyboard-first Interface**: Every action has a shortcut, mouse optional. - **Real-Time Monitoring**: Monitor progress from your phone. - **AI Arena**: Race agents head-to-head for a competitive view. ### Screenshots: - **Overview**: Multiple agents in parallel. - **Focus View**: Focused view on a single task. - **Diff Review**: Inline comments for easy review. - **AI Arena**: Race agents side by side. ### Why Parallel Code? - **Trustful Tools**: Runs Claude Code, Codex, and Gemini from one interface. - **Free and Open Source**: No extra subscription required. - **Isolation**: Keeps every change isolated and

### Overview: Motion Customization of Text-to-Video Diffusion Models: This project introduces MotionDirector, a tool designed to adapt existing text-to-video diffusion models to generate diverse videos with desired motions. The project is based on the research presented at ECCV 2024, aiming to provide a method for customizing text-to-video diffusion models to achieve specific motions. ### Key Features: - **Customizable Motions:** MotionDirector allows users to customize text-to-video diffusion models to generate videos with any desired motion. - **Versatile Applications:** The tool can be applied to various scenarios, including appearance customization and motion modification. - **User-Friendly Interface:** The project includes a user-friendly interface and detailed documentation to facilitate easy implementation and usage. ### Tech Stack: - **Python:** The primary programming language used for the project. ### Why It Matters: MotionDirector offers a powerful solution for generating videos with specific motions, making it valuable for various applications in media production, animation, and beyond. Its ability to adapt existing models to achieve desired results is particularly useful in creating consistent and high-quality videos with consistent motion. ### Repository: The repository for MotionDirector is hosted on GitHub, where it has been starred by over 1,000 users, indicating its growing interest and utility in the community. ### Primary Language: The primary language used in the project is Python. ### GitHub Stars: The repository has received 1,059 stars from the community, reflecting its popularity and the interest it has garnered. ### Demos: - **Demo Video:** A video showcasing the capabilities

Overview: This repository contains the implementation of the DreamerV2 agent, a world model agent that achieves human-level performance on the Atari benchmark. It includes training curves for all 55 games and is compatible with both discrete and continuous actions. Key features: - Training curves for all 55 games - Implementation of the DreamerV2 agent - Alternates between training the world model, policy, and collecting experience - Runs on a single GPU Tech stack: Python Why it matters: DreamerV2 is the first world model agent that achieves human-level performance on the Atari benchmark, surpassing the final performance of the top model-free agents Rainbow and IQN using the same amount of experience and computation.

Overview: This repository teaches how AI agents actually work by building a single agent step by step from a local LLM call, without frameworks, cloud APIs, or hidden reasoning. Key features: - Text in / text out - Roles and behavior (system prompts) - Structured output (JSON contracts) - Decisions (routing logic) - Tools (external capabilities) - Agent loop (observe → decide → act) - Memory (short and long-term) - Planning (as data, not thoughts) - Atomic actions (safe execution) - AoT - Atom of Thought (dependency graphs) - Evals (regression testing) - Telemetry (runtime observability) Tech stack: Python Why it matters: This repository provides a clean, local-first introduction to AI agents, emphasizing the underlying principles of how agents work without relying on frameworks, cloud APIs, or hidden reasoning. It is useful for developers, learners, and educators who want a deeper understanding of AI agent development.

### Overview: ONE-PEACE is a general representation model that excels across vision, audio, and language modalities. It achieves leading results without relying on pre-trained vision or language models, demonstrating strong zero-shot retrieval capabilities. ### Key Features: - **General Representation Model**: ONE-PEACE can represent and retrieve information across multiple modalities. - **Zero-Shot Retrieval**: It aligns modalities that are not paired in the training data, showcasing strong emergent zero-shot retrieval capabilities. - **Versatile Tasks**: It can be fine-tuned for various tasks such as vision, audio, and vision-language tasks. ### Tech Stack: - **Python**: The primary programming language used for development and training. ### Why It Matters: ONE-PEACE's ability to represent and retrieve information across multiple modalities without the need for pre-trained models makes it highly versatile and adaptable to various applications, including vision, audio, and language tasks. Its strong zero-shot retrieval capability further enhances its utility in unsupervised learning and unsupervised representation learning tasks.

### Overview: PointLLM is a multi-modal large language model designed to understand colored point clouds of objects. It can perceive object types, geometric structures, and appearance without considering ambiguous depth, occlusion, or viewpoint dependency. The model is trained on a novel dataset of 660K simple and 70K complex point-text instruction pairs to rigorously evaluate its perceptual abilities and generalization capabilities. ### Key features: - **Multi-modal Understanding**: PointLLM integrates text and image inputs to provide comprehensive understanding of point clouds. - **Object Type Recognition**: It accurately identifies object types from point clouds. - **Geometric Structure Analysis**: The model can analyze and describe geometric structures within point clouds. - **Appearance Understanding**: PointLLM can interpret the appearance of objects represented in point clouds. ### Tech stack: Python ### Why it matters: PointLLM is a significant advancement in the field of large language models, particularly in their ability to interpret and understand complex visual data such as point clouds. Its capabilities are crucial for applications in robotics, autonomous vehicles, and other fields where precise object recognition and understanding are essential.

Overview: Agentic Ra A security scanner for your LLM agentic workflows! Key features: - Security scanning for LLM workflows - Integration with various security tools - Detailed vulnerability reports Tech stack: Python Why it matters: Helps identify and mitigate security risks in your language model-driven workflows. Repository: splx-ai/agentic-radar Primary language: Python GitHub stars: 1058 GitHub description: A security scanner for your LLM agentic workflows

Overview: RepoAgent is an LLM-powered framework designed to assist developers and teams in generating repository-level code documentation quickly and efficiently. Key features: - Automatically detects changes in Git repositories, tracking additions, deletions, and modifications of files. - Independently analyzes the code structure through Abstract Syntax Tree (AST), generating documents for individual objects. - Accurately identifies inter-object bidirectional invocation relationships, enriching the global perspective of document content. - Seamlessly replaces Markdown content based on changes, maintaining consistency in documentation. - Executes multi-threaded concurrent operations, enhancing the efficiency of document generation. - Offers a sustainable, automated documentation update method for team collaboration. - Displays code documentation in an amazing way, with a document book per project powered by Gitbook. Tech stack: Python Why it matters: RepoAgent simplifies the process of creating and maintaining comprehensive project documentation, making it easier for developers to understand and modify their codebases, and ensuring the project remains accessible and modifiable over time. Repository: OpenBMB/RepoAgent GitHub stars: 1056 GitHub description: An LLM-powered repository agent designed to assist developers and teams in generating documentation and understanding repositories quickly.

### Overview: This repository contains the solution for the Zhihu Machine Learning Challenge 2017, where the primary language is Python. The project implements various text-classification models, including CNN, RNN (LSTM), RCNN, Inception, and FastText. The models are trained using PyTorch and the models path is provided in the `config.py` file. ### Key features: - **Models Implemented**: CNN, RNN (LSTM), RCNN, Inception, FastText. - **Training Methods**: Various training methods including without data augmentation and with data augmentation. - **Data Preprocessing**: The README provides detailed instructions on how to preprocess the data, including embedding files, question set, labels, and validation data. ### Tech stack: - **Python**: The primary programming language used. - **PyTorch**: The deep learning framework used for training the models. ### Why it matters: The project showcases the effectiveness of various text-classification models in handling different types of text data, demonstrating the versatility and performance of different architectures in machine learning tasks.

Overview: StoryToolkitAI is a film editing tool that utilizes AI to transcribe, index, and search footage, integrating with popular AI models like ChatGPT and others. Key features: - Full video indexing and search - Free automatic transcriptions on local machine - Free automatic translation to English on local machine - Compatible with OpenAI, ollama, vLLM, and DeepSeek - chat with AI about content or generate new ideas - Search content intuitively without typing exact words - Story editor for writing screenplays and exporting to EDL/XML/Fountain - Translate transcripts to other languages using OpenAI GPT - Ask AI to create stories and selections based on footage - Automatic speaker detection in transcripts - Project file management for intuitive workflows - Automatic question detection in transcripts - Multi-format export of transcripts, including SRT, TXT, AVID DS, and Fusion Text node - Import of existing SRT files - Easy copy of time-coded transcript text to clipboard Tech stack: Python Why it matters: StoryToolkitAI provides a powerful tool for film editors and content creators, enhancing their workflow with AI-assisted transcription, indexing, and search capabilities.