Projects

### Overview: Ouroboros is an open-source, self-creating AI agent that continues its identity, memory, and history across tasks and restarts. It runs as a native desktop app or through a headless CLI, with the runtime keeping its repository, durable memory, history, and interface on your machine. Model inference can use remote APIs or a local GGUF model. ### Key features: - **Self-creating AI agent**: Continues its identity, memory, and history across tasks and restarts. - **External projects support**: Works on external projects and coordinates a live swarm of specialist agents. - **Code rewrite capability**: Can rewrite the implementation it runs on, including its code, architecture, prompts, tools, and dependencies. - **Reflection and self-modification**: Reflection can change how it understands itself without severing that continuity. ### Tech stack: Python ### Why it matters: Ouroboros is an open-source, general-purpose AI agent that demonstrates the capabilities of self-creating AI agents, external project support, and code rewrite capability. It can be used in various applications, such as coordinating a live swarm of specialist agents or rewriting the implementation it runs on.

### Overview: StarNet is a local-first desktop harness that empowers developers to create and run real AI agents, transforming their workspace into a pixel-art station where agents perform real work with real models and tools. ### Key Features: - **A Crew, Not a Chatbot**: Create distinct agents with classes, personas, and loadouts. - **Bring Your Own Models**: Paste an OpenRouter key or sign in with supported providers. - **Message It from Anywhere**: Wire agents to popular chat apps for remote collaboration. - **Night Shift**: Leave the station running and agents keep working with explicit, adjustable leashes. - **Recipes, Skills, Schedules**: Launch recipes, grant skills, and run work on cron schedules. - **Asks Before It Guesses**: Turn ambiguity into concrete questions with options instead of silent wrong guesses. - **Finished Work Has a Front Door**: Deliverables land as real files, not chat scrollback. - **MCP Connectors**: Attach Model Context Protocol servers and paste-a-key/OAuth connectors. - **Voice**: Push-to-talk dictation in, one consistent station voice out. - **Real Ledgers**: Spend, budgets, and run history persist on disk and are shown as-is. ### Tech Stack: JavaScript ### Why It Matters: StarNet is a practical tool for developers who want to create and run real AI agents in a local-first environment. It transforms their workspace into a pixel-art station where agents perform real work with real models and tools, ensuring that the interface never asserts state the harness cannot prove.

### Overview: This repository is a comprehensive list of Implicit Representations and NeRF papers related to the Robotics/RL domain. It includes papers, codes, and related websites. ### Key features: - Papers - Codes - Websites ### Tech stack: Python, JavaScript, and Markdown ### Why it matters: This repository provides a curated list of papers, codes, and related resources for researchers and practitioners interested in Implicit Representations and NeRF techniques in the context of Robotics and Reinforcement Learning.

Overview: This repository contains the official codebase for the second edition of the book "Generative Deep Learning: Teaching Machines to Paint, Write, Compose and Play" by O'Reilly Media. Key features: - Implementations of various generative deep learning models - Training scripts for the book's chapters - Data download and preprocessing tools - Tensorboard monitoring for model training progress Tech stack: Jupyter Notebook Why it matters: This repository serves as a comprehensive resource for anyone studying generative deep learning, providing practical code examples and detailed explanations from the book's second edition.

Overview: ChatArena is a library designed to create multi-agent language game environments for Large Language Models (LLMs), focusing on developing communication and collaboration capabilities of AI agents. Key features: - Abstraction: Provides a flexible framework for defining multiple players, environments, and interactions using Markov Decision Process. - Language Game Environments: Offers a set of environments to help understand, benchmark, or train LLM agents. - User-friendly Interfaces: Includes both Web UI and CLI for developing and prompting LLM agents to act in environments. Tech stack: Python Why it matters: ChatArena is essential for researchers and developers interested in exploring the capabilities of LLMs in social interactions and communication, providing a robust platform for creating and testing AI agents in various language game scenarios.

### Overview: DepthFlow is a free and open-source image-to-video converter that transforms static images into stunning 3D parallax effect videos. It supports high-quality videos with seamless loops and fast processing, making it perfect for various applications. ### Key Features: - **High Quality Videos**: Rendered at any resolution, frame rate, and native supersampling using SOTA Depth Estimation models. - **Fast Processing**: Optimized Ray Marching GLSL Shader and TurboPipe for lightning-fast rendering, up to 8k50fps on an RTX 3060-class GPU. - **Commercial Use**: Encourages commercial use with a retribution policy, promoting fair use. - **Local Code**: No watermarks, unlimited usage, and scripts for mass production. ### Tech Stack: Python ### Why It Matters: DepthFlow offers a unique and high-quality solution for creating 3D parallax effect videos, perfect for digital art, social media, stock footage, and more. Its fast processing and commercial-friendly use make it a valuable tool for creators and developers alike.

Overview: RandLA-Net is an efficient neural architecture for semantic segmentation of large-scale 3D point clouds, presented at CVPR 2020. It offers a simple yet effective solution for handling large-scale point cloud data. Key features: - Efficient semantic segmentation - Works with large-scale 3D point clouds - Simple architecture for easy implementation Tech stack: Python, Tensorflow Why it matters: This project provides a practical solution for semantic segmentation of large-scale 3D point clouds, making it useful for applications requiring high accuracy and efficiency in processing large datasets. Repository: QingyongHu/RandLA-Net GitHub stars: 1565 GitHub description: 🔥RandLA-Net in Tensorflow (CVPR 2020, Oral & IEEE TPAMI 2021)

Overview: This repository contains a collection of outstanding papers and resources on large language model (LLM) related recommender systems, including surveys and updates on the latest research. Key features: - Survey papers on LLM in recommender systems - Latest research works updated regularly - Weekly paper notes on LLM-enhanced recommender systems Tech stack: Python, Markdown, LaTeX Why it matters: This repository provides valuable insights and resources for researchers and developers interested in the intersection of large language models and recommender systems, offering a comprehensive overview of the field and the latest advancements.

### Overview: Dolma is a comprehensive dataset and toolkit designed for pre-training large language models, including the Allen Institute for AI's (AI2) OLMo project. It comprises 3 trillion tokens from diverse sources and is available on the HuggingFace Hub under the ODC-BY license. ### Key Features: 1. **High Performance**: Capable of processing billions of documents concurrently. 2. **Portability**: Works on a single machine, a cluster, or cloud environments. 3. **Built-In Taggers**: Includes ready-to-use taggers for common data curation tasks. 4. **Fast Deduplication**: Uses a Rust Bloom filter for efficient document deduplication. 5. **Extensibility**: Supports custom taggers and AWS S3-compatible locations. ### Tech Stack: Python ### Why It Matters: Dolma is crucial for researchers and developers looking to enhance their language model training, providing a rich, diverse corpus that can significantly improve model performance and capabilities.

### Overview: This project is an official course on Large Language Models (LLMs) inference, focusing on the SGLang inference engine. The course covers understanding inference concepts, building a mini-SGLang from scratch, and finally diving into the real SGLang source code. It's designed for developers looking to enhance their skills in inference engines and contribute to open-source projects. ### Key Features: - **Understand Inference Concepts**: KV Cache, prefill/decode, compute-bound vs. memory-bound, and their implications. - **Build the Engine**: From scratch, build a mini-SGLang with every step in code. - **Read the Real Engine**: Dive into the implementation details of SGLang's cutting-edge optimizations. - **Touch the Frontier**: Explore advanced topics like quantization, hierarchical caching, DP Attention, Prefill-Decode disaggregation. - **Join Open Source**: Learn profiling and trace analysis, and contribute to open-source projects. ### Tech Stack: Python ### Why It Matters: This course fills the gap in tutorials on inference engines, providing a practical, hands-on approach to understanding and building inference engines. It's suitable for developers looking to enhance their skills in inference engines and contribute to open-source projects.

### Overview: OpenSeq2Seq is a toolkit designed to facilitate the exploration and experimentation with sequence-to-sequence models in the fields of speech recognition, text-to-speech, and natural language processing (NLP). It supports efficient distributed and mixed-precision training, making it ideal for researchers and developers looking to leverage the power of modern GPUs. ### Key Features: - **Models for:** Neural Machine Translation, Automatic Speech Recognition, Speech Synthesis, Language Modeling, and NLP tasks (sentiment analysis). - **Data-parallel Distributed Training:** Supports multi-GPU and multi-node setups. - **Mixed Precision Training:** Optimized for NVIDIA Volta/Turing GPUs, enhancing training efficiency. ### Tech Stack: Python ### Why It Matters: OpenSeq2Seq provides a comprehensive framework for researchers to experiment with various sequence-to-sequence models, leveraging the efficiency and scalability of distributed and mixed-precision training. This toolkit is essential for advancing research in speech processing, language understanding, and machine translation. ### Repository: GitHub: NVIDIA/OpenSeq2Seq Stars: 1560 Description: Toolkit for efficient experimentation with Speech Recognition, Text2Speech, and NLP ### Documentation: Documentation: https://nvidia.github.io/OpenSeq2Seq/ ### Software Requirements: - Python >= 3.5 - TensorFlow >= 1.10 - CUDA >= 9.0, cuDNN >= 7.0 - Horovod >= 0.13 (recommended for multi-GPU setup) ### Acknowledgments: - Speech-to-text workflow

### Overview: Agmsg is a cross-vendor messaging system designed for CLI AI agents. It allows Claude Code, Codex, Gemini, and GitHub Copilot to communicate directly with each other without the need for a central daemon or complex setup. The system uses a shared SQLite database and is entirely client-side, ensuring no human intervention and minimal complexity. ### Key Features: - **Cross-Agent Communication:** Agents can talk to each other directly through a shared SQLite database. - **No Daemon Required:** No need for a separate server or daemon process. - **No Network:** No external network connections are required. - **No Complexity:** Simplifies communication between AI agents without the overhead of a framework or daemon. ### Tech Stack: - **Shell:** Utilizes `bash` and `sqlite3` for its core functionality. - **SQLite:** Stores and manages the communication data in a SQLite database. ### Why It Matters: - **Simplicity:** Removes the need for a central server, making it easier to set up and manage. - **Flexibility:** Supports multiple AI agents without the need for a complex setup. - **Real-Time Communication:** Agents can communicate in real-time, eliminating the need for a human intermediary. ### Repository: - **GitHub Repository:** `fujibee/agmsg` - **Stars:** 1,560 - **Description:** Cross-agent messaging for CLI AI agents. No daemon required, no network, no complexity. ### Installation: - **Setup Script:** `setup.sh` provides a quick and easy way to install Agmsg. - **Clone

Overview: GIMP-ML is an AI-powered extension for the GNU Image Manipulation Program (GIMP), designed to integrate cutting-edge computer vision models into the GIMP interface for enhanced image manipulation capabilities. Key features: - GIMP-ML plugins for various image editing tasks using modern AI models. - Support for Dalle 2 and Dalle 3 for text-to-image and outpainting functionalities. - Updated code base to support newer and more stable functionality. - Updated models available in the repository. Tech stack: Python. Why it matters: GIMP-ML provides developers with a powerful toolset for advanced image editing tasks, leveraging the latest AI technologies to enhance the GIMP experience.

### Overview: Better Agents is a CLI tool and set of standards designed to enhance the performance and collaboration of coding assistants, making them experts in any agent framework and ready for production. ### Key Features: - **Standardized Structure**: `AGENTS.md` ensures every new feature is properly tested and evaluated, maintaining project maintainability. - **Versioned Prompts**: Versioned prompts for team collaboration, ensuring collaboration and performance evaluation. - **Scenario Tests**: End-to-end scenario tests for every feature, simulating agent behavior and ensuring expected outcomes. - **Prompt Registry**: Prompts are versioned and synced, allowing for easy collaboration and playgrounds. - **MCP Configuration**: MCP server configuration for expert coding assistants in your chosen framework. ### Tech Stack: TypeScript ### Why it Matters: Better Agents provides a structured approach to agent development, ensuring best practices and collaboration through standardized structure and prompt versioning, making your agents ready for production and easy to maintain. Repository: langwatch/better-agents GitHub Stars: 1559

Overview: Director is an AI framework designed to simplify video workflows by enabling developers, creators, and teams to integrate cutting-edge AI technologies into their media projects. Key features: - 20+ pre-built video agents for summarization, movie generation, indexing, organization, dubbing, and more. - A chat-based interface for video playback and intuitive controls. - Seamless integration with various GenAI APIs and databases. Tech stack: Python. Why it matters: Director leverages VideoDB's 'video-as-data' infrastructure, making it a powerful tool for developers looking to streamline their media workflows and unlock new possibilities with AI.

### Overview: This repository contains 228 battle-tested traps that map the footguns of Kotlin, Compose Multiplatform, and the desktop JVM development landscape. The traps are grouped into ten area skills, each providing specific ways a technique fails in practice, paired with a way to verify the failure and the fix on your own tree. The corpus is 28,000+ lines across 228 files, and in most of them, the largest section is **Traps**: the specific ways a technique fails in practice, each paired with a way to verify the failure and the fix on your own tree. ### Key features: - 228 battle-tested traps - 10 area skills in the open agent-skills format - 28,000+ lines across 228 files - Traps provide specific ways a technique fails in practice - Each trap is a standalone markdown file - Focus on failure modes and generic core that applies to any API ### Tech stack: Python ### Why it matters: This repository records what happened when its author shipped against the APIs for years, detailing what APIs report success while discarding data, the build flags that mean something different on one platform, the verification steps that pass on broken code, the migrations that deleted the wrong thing first. None of it is theory. Every lesson here is older than the documentation that describes it, each one having been paid for in crash reports, failed releases, review rounds, and user-filed issues before it was written down.

### Overview DeepMoji is a state-of-the-art deep learning model designed to analyze sentiment, emotion, and sarcasm in text. It leverages a large dataset of 1.2 billion tweets with emojis to achieve remarkable performance on various emotion-related tasks. ### Key Features - **Model Pretrained on Twitter Data**: DeepMoji is trained on a vast corpus of tweets containing emojis, enabling it to effectively identify and analyze emotional content. - **Transfer Learning**: The model can be fine-tuned on new datasets to adapt to specific applications, such as analyzing YouTube videos or social media posts. - **Vocabulary and Model**: Includes all necessary code to convert datasets, load the model, and run predictions. ### Tech Stack - **Python 2.7**: The primary language used for development. - **Keras Backend**: Requires either Theano or Tensorflow as the backend. ### Why It Matters - **State-of-the-Art Performance**: Achieves high accuracy in detecting emotions and sarcasm in text. - **Versatile Applications**: Can be applied to various domains, including social media analysis, sentiment classification, and more. ### Repository - **GitHub Stars**: 1555 - **Description**: A model trained on 1.2 billion tweets with emojis to understand how language is used to express emotions. It provides state-of-the-art performance on many emotion-related text modeling tasks. ### Installation - **Dependencies**: Requires Keras (tested on version 2.0.5), scikit-learn, h5py, text-unidecode

Overview: SALMONN is a family of advanced multi-modal large language models, including SALMONN 2, ELLSA 2, and others, designed for various applications such as speech quality assessment and audio-visual generation. Key features: - SALMONN 2: Unified vision, speech, text, and action in a streaming full-duplex framework. - ELLSA 2: First end-to-end model that unifies vision, speech, text, and action in a streaming full-duplex framework. - SALMONN: Audio-visual LLM for high-quality audio-visual video captions. - SALMONN 2: Reasoning-enhanced Audio-visual Large Language Model. - SALMONN 2: Captions and Reasoning for Video QA. - SALMONN 2: Speech Quality Assessment. Tech stack: Python, PyTorch, TensorFlow, and other relevant frameworks. Why it matters: SALMONN models provide advanced capabilities in multimodal understanding and generation, making them valuable tools for researchers and developers in various fields, including speech quality assessment, audio-visual generation, and more. Repository: bytedance/SALMONN GitHub stars: 1549 GitHub description: SALMONN family: A suite of advanced multi-modal LLMs.

Overview: This project introduces a method called SAM-Adapter, which adapts Meta AI's Segment Anything model to perform better on underperforming scenes. The project is based on the research presented at the IEEE/CVF International Conference on Computer Vision (ICCV). Key features: - SAM-Adapter can be used to improve performance in various tasks, including camouflage, shadow detection, and medical image segmentation. - It supports both SAM2 and SAM3 models, with SAM3 being more powerful. - The project includes a new technical report on adapting SAM2 in medical datasets. - The project provides pre-trained models and detailed documentation for training and evaluation. Tech stack: The project is implemented using Python and PyTorch. Why it matters: This method demonstrates the adaptability of the Segment Anything model and provides a practical solution for improving performance in underperforming scenes, particularly in medical applications. Repository: The project is hosted on GitHub at tianrun-chen/SAM-Adapter-PyTorch. GitHub stars: The repository has received 1556 stars, indicating high interest and engagement from the developer community. Primary language: The project is written in Python, making it accessible for developers familiar with the language. Environment: The project requires Python 3.8 and PyTorch 1.13.0, and users can install all requirements by running `pip install -r requirements.txt`. Quick Start: The project provides clear instructions for downloading datasets, pre-trained models, and training and evaluation methods. It also offers a way to gain faster training by using the `torchrun`

### Overview: This repository is designed to guide you through your entire AI journey in DevOps, from beginner tools to advanced infrastructure roles. Whether you're just starting out or looking to accelerate your career, this comprehensive set of resources and frameworks will help you succeed. ### Key Features: - **AI Roadmap Overview**: A visual guide through 3 phases of AI development. - **MCP for DevOps**: A guide to building MCP servers with Golang and Kubernetes. - **AI Agent for DevOps**: A guide to building AI agents with Golang and LangChain. - **AI Agent for Monitoring**: A guide to monitoring systems with the Versus SRE Agent. - **AI Project Management**: An Engineer's Journey to building an intelligent project management system. - **SRE Agent Brain**: A guide to building the SRE Agent's runbook brain for system knowledge. ### Tech Stack: Python ### Why it Matters: This repository provides a structured and comprehensive approach to mastering AI in DevOps, ensuring safe and effective implementation across teams and organizations.