Projects

Overview: Orbit is a Python package designed for Bayesian forecasting tasks, offering an intuitive interface for time series analysis. It utilizes probabilistic programming languages under the hood, supporting models like Exponential Smoothing, Local Global Trend, Damped Local Trend, and Kernel Time-based Regression. Key features: - Bayesian forecasting with object-oriented design - Probabilistic models for accurate predictions - Support for various sampling and optimization methods Tech stack: Python Why it matters: Orbit simplifies Bayesian forecasting by providing a user-friendly interface while leveraging advanced probabilistic programming techniques, making it valuable for researchers and practitioners in the field of time series analysis and machine learning. Repository: The repository is hosted on GitHub at [https://github.com/uber/orbit](https://github.com/uber/orbit). GitHub stars: 2083 GitHub description: A Python package for Bayesian forecasting with object-oriented design and probabilistic models under the hood.  ------------------------------------------------------------------------------------------------------------------------------------- <!--- BADGES: START --->  [][#pypi-package] [](https://github.com/uber/orbit/actions/workflows/test.yaml) [![Documentation Status](https://readthedocs.org

Overview: ComfyUI is an official front-end implementation of ComfyUI, providing a seamless user experience across various platforms. It includes features such as native translation and enhanced accessibility. Key features: - Native translation - Enhanced accessibility Tech stack: TypeScript Why it matters: ComfyUI offers a robust and user-friendly interface, making it an essential tool for developers working on front-end projects. Repository: Comfy-Org/ComfyUI_frontend Primary language: TypeScript GitHub stars: 2082 GitHub description: Official front-end implementation of ComfyUI README excerpt: Official front-end implementation of [ComfyUI](https://github.com/comfyanonymous/ComfyUI).

### Overview: Machine Learning Visualized is a Jupyter Book containing Jupyter Notebooks that implement and derive machine learning algorithms from first principles. It includes interactive notebooks built with Marimo, allowing visualization of how weights influence loss functions. The repository provides the code to configure and build the Jupyter Book, with the website updated via GitHub Actions after every commit or pull request. ### Key features: - Interactive notebooks built with Marimo - Visualizations of machine learning algorithms throughout their training phase - Code for configuring and building the Jupyter Book - Updated website using GitHub Actions ### Tech stack: TeX ### Why it matters: This repository is a comprehensive resource for understanding machine learning algorithms through visual and interactive means, making complex concepts accessible and engaging.

### Overview: A proof-of-concept Jupyter extension that converts English queries into relevant Python code. ### Key features: - Converts English queries into Python code - Supports Jupyter Notebook - Includes a demo video and blog post - Compatible with Ubuntu and macOS ### Tech stack: Python, Jupyter Notebook, Sentence Transformers ### Why it matters: This extension simplifies data analysis by directly translating English queries into Python code, making it easier to perform complex data manipulations and analyses. ### Repository: deepklarity/jupyter-text2code ### Installation: - Clone the repository - Install dependencies - Enable Jupyter extension - Run Jupyter notebook server ### Uninstallation: - Uninstall the extension ### Supported Operating Systems: - Ubuntu - macOS ### Docker containers: - CPU image: 1.51 GB - GPU image: 2.56 GB ### Model Training: - Supports pandas commands and snippets from awesome-notebooks - Uses SentenceTransformers `paraphrase-MiniLM-L6-v2` for better performance ### Steps to add more intents: - Add templates - Generate training data - Train intent index - Modify plugin code - Reinstall plugin ### TODO: - Add Ollama support - Publish Docker image - Refactor code - Add support for more commands - Improve intent detection and NER - Bind with small LLMs - Explore sentence paraphrasing - Gather variable names and library names - Train language model for direct English->code

Overview: Snntorch is a Python package that extends the capabilities of PyTorch to perform gradient-based learning with spiking neural networks. It seamlessly integrates with PyTorch, leveraging GPU acceleration for efficient computation and treating spiking neurons as recurrent activation units. Key features: - A spiking neuron library similar to PyTorch's nn library - Enables exporting to other SNN libraries via the NIR format - Provides common arithmetic operations on spikes - A library for spike generation and data conversion - Visualization tools for spike-based data - Optional surrogate gradient functions - Intuitive use with PyTorch Tech stack: Python Why it matters: Snntorch is essential for researchers and developers interested in creating more efficient neural networks inspired by the brain's spike-based encoding mechanism. It simplifies the implementation of spiking neural networks in Python, making it easier to experiment with and integrate into existing projects. Repository: jeshraghian/snntorch GitHub stars: 2078 GitHub description: Deep and online learning with spiking neural networks in Python README excerpt: .. image:: https://github.com/jeshraghian/snntorch/blob/master/docs/_static/img/spike_excite_alpha_ps2.gif?raw=true :align: center :width: 800 If you like this project, please consider starring ⭐ this repo as it is the easiest and best way to support it. If you have issues or comments, or are looking for advice on training spiking neural networks, you can open an issue or a

Overview: AICI is a powerful tool that lets you build and experiment with controllers for large language models (LLMs) to enhance their output. It simplifies the process of writing and deploying controllers, making it easier to implement diverse strategies and improve LLM generations. Key features: - Custom logic during token-by-token decoding - State management during LLM requests - Flexible implementation in any language that compiles to WebAssembly (Wasm) or can be interpreted inside Wasm - Secure sandboxing to prevent access to filesystem, network, or other resources - Fast performance with minimal overhead on the token generation process Tech stack: Rust, WebAssembly (Wasm) Why it matters: AICI simplifies the development of controllers for LLMs, making it easier to implement diverse strategies and improve LLM generations. It abstracts away implementation details, simplifying the development process and making it easier to write fast controllers. Repository: microsoft/aici GitHub stars: 2076 GitHub description: AICI: Prompts as (Wasm) Programs

Overview: This repository contains a simple interface for editing natural photos using generative neural networks, specifically the paper "Neural Photo Editing with Introspective Adversarial Networks." Key features: - A user-friendly interface for editing photos - Support for a variety of generative neural network models - Compatibility with different operating systems Tech stack: Python, Theano, Lasagne, cuDNN Why it matters: This project provides a practical tool for editing and enhancing natural photos using advanced neural network techniques, making it useful for both researchers and developers looking to explore generative AI in the field of image editing.

### Overview: Universal Data Tool is a powerful web and desktop application designed for easy collaboration and annotation of various types of data, including images, text, audio, and documents. It streamlines the process of creating, labeling, and managing datasets, making it accessible and efficient for both teams and individuals. ### Key Features: - **Real-time Collaboration**: Seamlessly collaborate with others in real time without the need for sign-up. - **Extensible Data Formats**: Supports the .udt.json and .udt.csv formats, allowing for easy integration with any dataset. - **Web/Desktop Compatibility**: Available as a web app and desktop app, ensuring flexibility and accessibility. - **Image Segmentation and Classification**: Ideal for image-based datasets, including segmentation and classification tasks. - **Text Classification, Named Entity Recognition, and Part of Speech Tagging**: Perfect for text-based datasets. - **Audio Transcription**: Transcribe audio data for accurate transcription. - **Data Entry**: A comprehensive tool for data entry and management. - **Video Segmentation and Landmark Annotation**: Useful for video-based datasets. - **Extensive Dataset Support**: Supports a wide range of datasets, including image segmentation, classification, text classification, named entity recognition, part of speech tagging, audio transcription, data entry, video segmentation, and landmark annotation. ### Tech Stack: The Universal Data Tool is built using JavaScript, making it accessible and compatible with modern web technologies. ### Why It Matters: Universal Data Tool simplifies the process of creating, labeling, and managing datasets. Its real-time collaboration feature makes it ideal for teams,

Overview: Nodeterm is a node-based terminal manager designed for users with ADHD and scattered workflows. It provides a spatial layout of terminals and agents as draggable nodes on an infinite pan/zoom canvas, allowing projects to be visualized as a Trello-style board of live Claude Code sessions. Key features: - Multiple real terminals live as draggable nodes on a single pan/zoom canvas. - Projects are visualized as a Trello-style board of live Claude Code sessions. - Agents and agents' context can be linked to sticky notes. - Sessions are spatial and persistent, ensuring mental model integrity across restarts. Tech stack: TypeScript Why it matters: Nodeterm offers a unique and intuitive way to manage multiple terminal sessions and agents, providing a spatial and persistent workspace that enhances productivity and reduces workflow clutter. Repository: eneskirca/nodeterm Primary language: TypeScript GitHub stars: 2074

### Overview: ADR secures enterprise AI agents by detecting, observing, benchmarking, detecting, and preventing unsafe AI actions. Deployed at Uber, it helps organizations secure their AI agents in production. ### Key features: - **ADR Discovery:** Inventory AI apps, CLI agents, IDE extensions, model runtimes, and MCP servers on endpoints. - **ADR Observability:** Captures agent telemetry from various AI tools and support agents. - **ADR Benchmark:** Tests agent security under realistic enterprise conditions. - **ADR Detection:** Detects risky agent behavior efficiently using two-tier architecture. - **ADR Prevention:** Prevents unsafe actions before they cause harm. ### Tech stack: Python ### Why it matters: ADR secures enterprise AI agents by providing comprehensive security measures, making it essential for organizations looking to protect their AI agents in production.

### Overview: FISSURE is an open-source framework designed for RF analysis, automation, and distributed operations. It connects SDR hardware, signal processing, sensor nodes, geolocation, protocol analysis, targeting, and situational awareness within a single extensible environment. ### Key Features: - **Signal Analysis:** Move from survey and detection through capture, inspection, conditioning, feature extraction, classification, and protocol discovery. - **Targets & Actions:** Manage Targets and execute reusable plugin Actions through focused, sequential, fuzzing, and packet-crafting workflows. - **Sensor Nodes:** Configure local and remote nodes, manage hardware, transfer files, automate startup behavior, and monitor node status. - **Library:** Maintain reusable protocol information, packet definitions, archived signals, replay content, datasets, and RF reference material. - **Plugins & Actions:** Add, deploy, and execute capabilities across supported nodes and interfaces without modifying the FISSURE core. ### Tech Stack: Python ### Why it Matters: FISSURE is designed for experimentation, research, education, capability development, and operational integration without locking users to a single protocol, sensor, platform, or use case. It provides a unified environment for RF analysis, automation, and distributed operations, making it suitable for various applications in the field of RF and reverse engineering.

Overview: ElatoAI is a real-time voice AI platform that leverages 100+ models on Arduino ESP32, providing secure WebSocket and Edge Functions for seamless AI interactions. Key features: - Real-time voice AI models powered by 100+ models - Secure WebSocket and Edge Functions for global deployment - Supports over 100+ voice pipeline systems - Deployable on ESP32 devices for local AI models and voice generation Tech stack: TypeScript, Deno Edge Why it matters: ElatoAI simplifies the integration of AI into IoT devices, enabling them to understand and respond to voice commands in real-time, making them more interactive and intelligent.

Overview: The AGI repository is a living research project featuring a distributed AGI system that enables intelligence to compound continuously as more agents join. It uses peer-to-peer networking and fully autonomous AI agents to collaborate on experiments, share findings, and push breakthroughs. Key features: - Distributed inference: Queries route to the best model loaded across the network. - Shared providers: Members can pool OpenRouter / Groq / Together keys with per-member budgets. - Pod VM: Always-on agent daemon across multiple providers. - Pod Capsule: Portable `.tar.gz` of full pod state with AES-256-GCM encryption for seamless migration or self-hosting via `docker compose up`. - Distributed training: 32 anonymous nodes trained a language model across independent consumer devices in 24 hours. - Blockchain: Mysticeti consensus, stateless execution, and live economy with 695+ agents. Tech stack: JavaScript. Why it matters: This project showcases the potential of distributed AGI systems, highlighting continuous intelligence compounding and the benefits of peer-to-peer networking in accelerating research and development. Repository: hyperspaceai/agi GitHub stars: 2072. GitHub description: The first experimental distributed AGI system. Thousands of autonomous AI agents collaboratively train models, share experiments via P2P gossip, and push breakthroughs here. Fully peer-to-peer. Join from your browser or CLI.

### Overview: 本项目是一款基于深度学习的中文词段分割模型，采用BiLSTM+CRF和IDCNN+CRF两种架构进行训练。模型在待字闺中公众号获取语料下载地址，通过词向量和训练语料构建词表，最终生成词性标注模型。该模型在识别中文词段和公司名识别方面表现优异，已在实际应用中取得了显著效果。 ### Key features: - BiLSTM+CRF架构词段分割模型 - IDCNN+CRF架构词段分割模型 - 自定义词典支持 - 词性标注模型训练 - 实时Web服务接口 ### Tech stack: C++、Bazel、TensorFlow、Word2Vec ### Why it matters: 该模型在中文词段识别和公司名识别方面具有较高的准确率，广泛应用于文本处理和自然语言处理领域，具有重要的应用价值。

Overview: LlamaDeploy is a tool designed to deploy and manage agent workflows in production environments. Key features: - Simplified workflow deployment - Automated scaling and load balancing - Integration with various agent frameworks Tech stack: Python Why it matters: LlamaDeploy streamlines the deployment process, ensuring that workflows run smoothly and efficiently in production environments.

### Overview RoboSat is an end-to-end pipeline for feature extraction from aerial and satellite imagery. It extracts features such as buildings, parking lots, roads, water, and clouds. The pipeline is written in Python and comes with tools for data preparation, training segmentation models, and post-processing to clean and simplify the results. It works with the Slippy Map tile format to abstract away geo-referenced imagery. ### Key Features - **Data Preparation**: Tools for creating datasets from aerial imagery and OpenStreetMap geometries. - **Training Models**: Fully convolutional neural nets for segmentation. - **Post-Processing**: Cleaning and simplifying segmentation results into simple geometries. ### Tech Stack - Python 3 ### Why It Matters RoboSat is useful for developers and researchers interested in semantic segmentation of aerial and satellite imagery. It provides a comprehensive solution for extracting meaningful features from images, which can be applied in various applications such as urban mapping, infrastructure identification, and more.

### Overview: Superglue is a powerful AI platform designed to connect, migrate, and implement enterprise systems. It seamlessly integrates with various systems, including NetSuite, Sage Intacct, SAP, Business Central, and Acumatica, as well as connects with AI platforms like Claude. The platform automates the implementation process, reducing the need for human coordination and engineering work, making it ideal for businesses of all sizes. ### Key Features: - **NetSuite Migration:** Quickly migrate Sage Intacct, NetSuite, SAP, Business Central, and Acumatica systems. - **ERP & CRM Integration:** Connect ERP, CRM, databases, and internal systems to AI platforms. - **Customer Onboarding:** Implement customer systems, import historical data, and get integrations live in hours. - **End-to-End Implementation:** Manage the entire implementation process from start to finish. ### Tech Stack: Superglue is built on TypeScript, leveraging the latest AI and cloud technologies to ensure seamless integration and efficient implementation. ### Why It Matters: Superglue streamlines the implementation process, reducing the need for manual labor and improving the overall efficiency of system integration. Whether you're migrating legacy systems or setting up new integrations, Superglue ensures a smooth and effective process, making it an essential tool for businesses looking to modernize their systems and improve their operations. ### Repository: The repository for Superglue is hosted on GitHub, where it has received 2,066 stars. It is a cloud-hosted solution that can be deployed on your own infrastructure, ensuring

### Overview: OpenDAN is an open-source Personal AI Operating System designed to consolidate various AI modules for personal use, providing a seamless integration of diverse AI functionalities. ### Key features: - **Rapid Installation and Deployment**: OpenDAN is compatible with a wide range of hardware environments (PC, Mac, Raspberry Pi, NAS) through Docker. - **AI Agent Switching**: Supports locally running open-source models (LLaMa) for a large language model. - **Built-in AI Agents**: Personal Assistant Jarvis, Information Assistant Mia, Private English Teacher Tracy, and ai_bash (for developers). - **Connectivity**: Telegram/Email connectivity. - **Knowledge Base**: Local private Knowledge Base using existing files or emails. - **Workflow Integration**: Built-in Workflow story_maker for creating audio fairy tale books. - **Distributed AI Computing Core**: Supports complex selections. - **Manual Installation**: Agents/Workflows/Models can be manually downloaded and installed. ### Tech stack: Python ### Why it matters: OpenDAN is revolutionizing the AI landscape by providing a personal AI operating system that integrates diverse AI modules, ensuring seamless integration and control over AI agents.

### Overview: This project is a collection of LaTeX graphics examples for academic use, focusing on Bayesian networks, graphical models, tensors, and technical frameworks. It includes detailed illustrations that are easy to follow on Overleaf, making it ideal for researchers and educators in various scientific fields. ### Key features: - **Bayesian Networks**: Examples of Bayesian networks using different factorization methods. - **Graphical Models**: Illustrations of partially observed time series, matrix and tensor time series prediction, undirected graphs, and Laplacian convolutional time series modeling. - **`pgfplots` Function Plotting**: Examples of PDF and CDF functions, joint PDF functions of lognormal distributions, and iterative processes for solving linear equations. - **`tikz` for Matrix Structure**: Graphical illustrations of multivariate time series forecasting and temporal matrix factorization. - **`tikz-3dplot` for Tensor Structure**: Illustrations of rolling time series forecasting with temporal matrix factorization. ### Tech stack: TeX ### Why it matters: LaTeX is a powerful tool for creating high-quality graphics, especially in scientific and academic fields. This project provides a collection of examples that can be easily integrated into LaTeX documents, enhancing the presentation of complex data and models.

### Overview: DeePMD-kit is a deep learning package designed for many-body potential energy representation and molecular dynamics simulations. It allows you to start from a pretrained model, fine-tune it for your specific system, and deploy it at scale. ### Key Features: - Download pretrained DPA4 models and fine-tune full models. - Design and train models from scratch for new systems. - Support for modern model portfolios including DPA4 and DPA4C. - Flexible backend support for training and running models. - Capable of generating various properties like virials, Hessians, magnetic forces, and more. ### Tech Stack: Python ### Why It Matters: Using DeePMD-kit, you can leverage pretrained models as your starting point, fine-tune them for specific applications, and scale your simulations from small systems to large distributed training and molecular dynamics. This makes it a versatile tool for both academic and industrial research in materials science and molecular dynamics.