Projects

Overview: Foolbox is a Python library designed to help developers create adversarial examples that can fool neural networks using PyTorch, TensorFlow, and JAX. It provides a collection of state-of-the-art adversarial attacks and is built on top of EagerPy for native performance. Key features: - Native performance on models developed in PyTorch, TensorFlow, and JAX - Large collection of state-of-the-art gradient-based and decision-based adversarial attacks - Type checking for catching bugs before running code Tech stack: Python Why it matters: Foolbox simplifies the process of creating adversarial examples, making it easier for developers to benchmark the robustness of their machine learning models without the need for extensive setup or installation. Repository: bethgelab/foolbox GitHub stars: 2983

Overview: Reloadium is an advanced tool that provides hot reloading, profiling, and AI features for Python developers, enhancing productivity and debugging efficiency. Key features: - Hot Reloading: Real-time code updates without restarting the application. - Profiling: Detailed performance analysis and optimization. - AI Integration: Seamless integration with ChatGPT for improved AI responses. - Multi-File Reloading: Tracks and reloads dependencies across the entire project. - Django Integration: Hot reloading views and page refreshes. - Flask Integration: Hot reloading Flask apps and page updates. - SqlAlchemy Integration: Rollbacks to prevent unwanted database changes. - Pandas Integration: Facilitates hot reloading of Pandas objects. Tech stack: Python Why it matters: Reloadium simplifies the development process by providing real-time feedback, improved debugging, and seamless integration with popular frameworks and libraries, making it an essential tool for Python developers. Repository: reloadware/reloadium GitHub stars: 2982 GitHub description: Hot Reloading and Profiling for Python

Overview: Comic Translate is an AI-powered app and browser extension designed to automatically translate comics, manga, manhwa, BDs, fumetti, and other formats into multiple languages and formats. Key features: - Supports translating comics from English, Korean, Japanese, French, Simplified Chinese, Traditional Chinese, Russian, German, Dutch, Spanish, and Italian. - Automatically detects and translates text within images, PDFs, EPUBs, and other formats. - Provides a user-friendly interface for both desktop applications and browser extensions. Tech stack: Python Why it matters: Comic Translate leverages the capabilities of state-of-the-art Large Language Models (LLMs) to accurately translate comics across various languages and formats, offering a valuable tool for language learners and enthusiasts. Repository: ogkalu2/comic-translate GitHub stars: 2980 GitHub description: AI comic and manga translator app/browser extension for automatically translating comics, manga, manhwa, BDs, fumetti, and more in multiple languages and formats (Images, PDF, EPUB, CBR, CBZ etc).

Overview: AutoFlow is a Graph RAG-based knowledge base tool built on TiDB Serverless Vector Storage, designed for advanced conversational search and integration with LlamaIndex and DSPy. Key features: - Live Demo: [https://tidb.ai](https://tidb.ai?utm_source=tidb.ai&utm_medium=community) - Embeddable JavaScript Snippet: Integrate our conversational search window into your website - Deploy with Docker Compose: Requires 4 CPU cores and 8GB RAM Tech stack: TiDB, LlamaIndex, DSPy, Next.js, Tailwind CSS, shadcn/ui Why it matters: AutoFlow is an open-source tool for building conversational knowledge bases, leveraging advanced RAG techniques to enhance search capabilities and integration with popular AI models.

Overview: A comprehensive quantitative trading system that leverages AI and ML technologies for advanced factor analysis, real-time data processing, and risk management. Key features: - Multi-source data integration - Real-time market data streaming - Advanced factor models (momentum, value, quality, volatility) - Sentiment analysis using OpenAI GPT - Extensible strategy framework with 8+ built-in strategies - C++ core engine for low-latency order execution - Interactive dashboards with Plotly and Streamlit - Comprehensive risk controls and portfolio management Tech stack: Python Why it matters: This system provides a robust and production-ready solution for quantitative trading, combining traditional financial analysis with cutting-edge AI/ML technologies to enhance market insights and risk management. Repository: 0xemmkty/QuantMuse GitHub stars: 2977 GitHub description: A comprehensive quantitative trading system with AI-powered analysis, real-time data processing, and advanced risk management.

Overview: OpenVSLAM is a versatile Visual SLAM (Simultaneous Localization and Mapping) framework designed for developers. Key features: - Real-time localization and mapping - Support for various camera types - Robust performance in challenging environments - Open-source, community-driven development Tech stack: C++, Python, OpenCV Why it matters: OpenVSLAM provides a powerful and flexible solution for developers looking to integrate advanced localization and mapping capabilities into their projects, making it an essential tool for anyone working in robotics, augmented reality, or autonomous systems.

Overview: This project evaluates the performance of neural networks for object recognition using the mean Average Precision (mAP) metric. Key features: - Calculates Average Precision (AP) for each class - Computes the mean Average Precision (mAP) across all classes Tech stack: Python Why it matters: A higher mAP value indicates better performance of the neural net, as measured by the ground-truth and set of classes. Repository: Cartucho/mAP Primary language: Python GitHub stars: 2970 GitHub description: mean Average Precision - This code evaluates the performance of your neural net for object recognition. README excerpt: # mAP (mean Average Precision) [](https://github.com/Cartucho/mAP) This code will evaluate the performance of your neural net for object recognition. <p align="center"> <img src="https://user-images.githubusercontent.com/15831541/37559643-6738bcc8-2a21-11e8-8a07-ed836f19c5d9.gif" width="450" height="300" /> </p> In practice, a **higher mAP** value indicates a **better performance** of your neural net, given your ground-truth and set of classes. ## Citation This project was developed for the

### Overview: Ax is a TypeScript-first framework designed for building with Large Language Models (LLMs) across multiple ecosystems including TypeScript, Python, Java, C++, Go, and Rust. It provides a unified programming model that allows developers to create signatures, named deployment profiles, agents, flows, and optimizers that can be compiled into verified libraries for different runtimes. ### Key Features: - **Signature for Typed Structured Generation**: Supports string DSL, fluent `f()` builder, and Standard Schema v1 validators like Zod, Valibot, and ArkType. - **Named Deployment Profiles**: Enables deployment profiles across native APIs, routers, hosted inference, configurable runtimes, audio, and real-time event streams. - **Agents with Runtime Execution**: Agents execute with context budgets, checkpoints, action-log replay, discovery, memory, skills, and delegation. - **Flows as Typed Program Graphs**: Flows are typed program graphs with branches, loops, feedback, cache behavior, parallel execution, and `.returns(...)` projection. - **Optimizers**: Includes GEPA, few-shot bootstrapping, portable optimizer artifacts, and evaluation/apply flows. - **One Semantic Core**: Compiled into TypeScript, Python, Java, C++, Go, and Rust library shapes, ensuring the same Ax program model can move across runtime stacks. ### Tech Stack: Ax is TypeScript-first and ships as `@ax-llm/ax`. It compiles into verified generated Python, Java, C++, Go, and Rust libraries. ### Why It Matters: Ax simplifies the process of integrating L

### Overview: Minkowski Engine is an auto-differentiation library designed for sparse tensors, enabling efficient computation of neural network layers on high-dimensional data. ### Key Features: - Supports convolution, pooling, unpooling, and broadcasting operations for sparse tensors. - Provides CUDA acceleration for coordinate management functions. - Compatible with PyTorch 1.8 and 1.9, with Docker installation instructions available. ### Tech Stack: Python ### Why it Matters: Minkowski Engine simplifies the process of building and optimizing neural network architectures for sparse data, making it particularly useful in applications requiring high computational efficiency and reduced memory usage.

Overview: This repository provides an introduction to Deep Neural Networks using Keras and TensorFlow, covering topics such as artificial neural networks, deep learning frameworks, supervised learning, unsupervised learning, recurrent neural networks, and additional materials. Key features: - Introduction to Artificial Neural Networks - Introduction to Deep Learning Frameworks - Overview of Keras Layers - FC and MNIST - CNN and MNIST - VGG16, VGG19, ResNet50 - AutoEncoders and MNIST - LSTM for Sentence Generation - Custom Layers in Keras - Multi-modal Network Topologies with Keras Tech stack: Python, Keras, TensorFlow Why it matters: This tutorial is designed for developers looking to understand and implement deep learning concepts using Keras and TensorFlow, providing practical examples and explanations for various neural network architectures and techniques.

Overview: The J-Space Cognition Suite is a model-agnostic inference-time control suite designed for complex reasoning, repository engineering, coordinated agents, and authorized security analysis. It leverages Anthropic's J-space global workspace research to provide a robust framework for long-horizon work, verification, and recovery. Key features: - Supports multiple modules sharing a single premise and routing entry. - Uses standard-library Python scripts to persist state, reread source text, detect stale maps and evidence, and return blocking results when necessary. - Provides tools and agents for host management and skill execution. Tech stack: Python Why it matters: The J-Space Cognition Suite offers a scalable and flexible solution for complex reasoning tasks, making it suitable for real-world engineering and production environments. Its modular design and robust control mechanisms ensure reliable and efficient execution, even in high-stakes scenarios. Repository: Tiger3807861189/J-Space-Cognition-Suite GitHub stars: 2967 GitHub description: J-Space Cognition Suite — a model-agnostic inference-time control suite for complex reasoning, repository engineering, coordinated agents, and authorized security analysis. Based on Anthropic's J-space global workspace research.

### Overview: Infinity is a high-throughput, low-latency REST API designed for serving text-embeddings, reranking models, clip, clap, and colpali. The project utilizes the MIT License and supports packed embedding tensors with shape and dtype metadata, including token matrices for late interaction. It leverages dynamic batching and tokenization dedicated in worker threads for efficient inference. ### Key Features: - **Deploy any model from HuggingFace**: Supports embedding, reranking, clip, and sentence-transformer models. - **Fast inference backends**: Built on PyTorch, optimum (ONNX/TensorRT), and CTranslate2, utilizing FlashAttention for optimal performance on NVIDIA CUDA, AMD ROCM, CPU, AWS INF2, and Apple MPS accelerators. - **Multi-modal and multi-model**: Mix-and-match multiple models, orchestrating them. - **Tested implementation**: Unit and end-to-end tested, ensuring correct embeddings. - **Easy to use**: Built on FastAPI, allowing launching of all arguments via environment variables or command-line arguments. OpenAPI aligned to OpenAI's API specs, with a documentation accessible at [https://michaelfeil.github.io/infinity](https://michaelfeil.github.io/infinity/). ### Tech Stack: - **Python**: The project is built using Python, with packages utilizing the MIT License. - **FastAPI**: The backend is built on FastAPI, a modern web framework for building APIs. - **PyTorch**: The inference server is built on PyTorch, utilizing FlashAttention for optimal performance.

Overview: Databerry is a no-code platform designed for building custom Language Models for Agents (LLM Agents). Key features: - Instantly create custom LLM Agents - Intuitive drag-and-drop interface - Seamless integration with various tools and APIs - Easy deployment and management Tech stack: React, Node.js, and various LLM models Why it matters: Databerry simplifies the process of creating custom LLM Agents, making it accessible to developers without extensive coding experience.

Overview: This repository provides guidelines and best practices for publishing research code in Machine Learning, based on analysis of more than 200 repositories. It includes a README template and a checklist to ensure reproducibility and popularity. Key features: - A README template for ML research repositories - A checklist of items to check for completeness - Data and notebook for reproducing the analysis Tech stack: Python, Markdown, LaTeX Why it matters: The checklist helps ensure that research code is complete and easy to reproduce, leading to higher GitHub stars and better community reception.

### Overview: MITIE is a comprehensive library and toolkit for information extraction, providing state-of-the-art tools for named entity extraction and binary relation detection. It is built on top of dlib and utilizes advanced techniques like distributional word embeddings and Structural Support Vector Machines. ### Key features: - **Named Entity Extraction (NER)**: Performs advanced named entity recognition. - **Binary Relation Detection (BRD)**: Detects binary relations between entities. - **Training Custom Extractors and Detectors**: Allows users to train custom extractors and detectors. - **Bindings**: Available in C++, Python, R, Java, C, and MATLAB. - **Pre-trained Models**: Supports English, Spanish, and German models trained on various linguistic resources. ### Tech stack: MITIE is written in C++ and utilizes dlib for high-performance machine learning. ### Why it matters: MITIE offers powerful tools for information extraction, making it essential for researchers, developers, and businesses looking to enhance their data analysis capabilities. Its state-of-the-art techniques and pre-trained models make it a valuable resource in the field of natural language processing. ### Repository: The repository is hosted on GitHub at [https://github.com/mit-nlp/MITIE](https://github.com/mit-nlp/MITIE). ### GitHub stars: The repository has been starred by over 2,962 developers, indicating high interest and adoption. ### Primary language: The project is primarily written in C++. ### GitHub description: MITIE: MIT Information Extraction provides free, even for commercial use, state-of-the-art information extraction tools. It

### Overview: This repository contains a collection of highly cited and impactful papers related to Artificial Intelligence (AI), machine learning (ML), deep learning (DL), reinforcement learning (RL), and their applications. The repository includes a wide range of topics such as AI hardware, AI applications, game theory, deep learning, fairness and ethics in AI, machine learning, learning theory, and more. The papers are sourced from reputable journals and conferences, making them valuable resources for researchers and practitioners in the field. ### Key Features: - [AI Hardware](https://github.com/tirthajyoti/Papers-Literature-ML-DL-RL-AI/tree/master/AI%20Hardware) - [Application of Artificial Intelligence](https://github.com/tirthajyoti/Papers-Literature-ML-DL-RL-AI/tree/master/Application%20of%20AI) - [Artificial Intelligence (AI), and Game Theory](https://github.com/tirthajyoti/Papers-Literature-ML-DL-AI/tree/master/AI-Game-Theory) - [Deep Learning](https://github.com/tirthajyoti/Papers-Literature-ML-DL-AI/tree/master/Deep-learning) - [Fairness, Bias, and Ethics in AI](https://github.com/tirthajyoti/Papers-Literature-ML-DL-RL-AI/tree/master/Fairness%2C%20Bias%2C%20Ethics%20in%20AI%20and%2

Overview: This is a C++ implementation of ChatGLM-6B, ChatGLM2-6B, ChatGLM3, GLM-4, and CodeGeeX2, designed for real-time chatting on macOS, Linux, and Windows. Key features: - Pure C++ implementation based on ggml, compatible with llama.cpp - Accelerated memory-efficient CPU inference with int4/int8 quantization, optimized KV cache, and parallel computing - P-Tuning v2 and LoRA finetuned models support - Streaming generation with typewriter effect - Python binding, web demo, API servers, and more possibilities Tech stack: C++ Why it matters: This implementation is optimized for real-time chatting, making it suitable for applications requiring high performance and low latency. Repository: li-plus/chatglm.cpp GitHub stars: 2958

### Overview: Qwen Audio Agent is a real-time voice runtime designed to keep AI agents engaged and present during conversations. It ensures that agents remain active and responsive, enhancing the user experience by maintaining continuous interaction. ### Key Features: - **Real-time Voice Interaction:** Enables seamless communication between users and AI agents. - **Voice Frontends:** Supports various voice and video models for improved user engagement. - **Desktop Conversation Panel:** A dedicated panel for desktop users, providing real-time updates and progress tracking. - **Remote Access:** Allows agents to be accessed remotely, enhancing flexibility and accessibility. - **Customer Service:** Employs smart cockpit and digital human features for enhanced interaction. - **Examples and Scenario Expansion:** Expands on the examples provided, including customer service, smart cockpit, and digital human scenarios. ### Tech Stack: - **JavaScript:** The primary programming language used for development. - **Node.js:** Framework for building the orchestration runtime. - **DeepSeek Harness:** Integration with a deep-seek harness for native ACP support. - **Voice and Video Models:** Supports a variety of voice and video models for improved interaction. - **DeepSeek Harness:** Native ACP integration for improved performance and reliability. ### Why It Matters: Real conversation should not leave you waiting after a single sentence, nor should it grind to a halt just because the agent is looking something up, calling a tool, or working on a task. The Qwen Audio Agent ensures that conversation flows smoothly, keeping the agent always present and engaged.

### Overview: Torchaudio is a powerful library for audio signal processing built on top of PyTorch. It offers a wide range of tools and functions for processing audio data, making it an essential tool for machine learning applications in the audio domain. ### Key Features: - **Dataloaders for Common Audio Datasets:** Provides convenient ways to load and preprocess audio datasets. - **Audio and Speech Processing Functions:** Includes functions for forced alignment, spectrogram generation, and more. - **Common Audio Transforms:** Offers a variety of transforms such as Spectrogram, AmplitudeToDB, MelScale, and more. - **Compliance Interfaces:** Ensures compatibility with other libraries, making it easier to integrate with existing systems. ### Tech Stack: Python ### Why it Matters: TorchAudio simplifies the process of applying PyTorch to the audio domain, leveraging PyTorch's strengths in GPU acceleration, autograd, and consistent tensor operations. This makes it highly suitable for machine learning applications in audio processing, such as speech recognition, audio classification, and more. ### Repository: GitHub Repository: `pytorch/audio` GitHub Stars: 2957 ### Installation: ```bash pip install torchaudio ``` ### API Reference: API Reference is located [here](http://pytorch.org/audio/main/). ### Contributing Guidelines: Please refer to [CONTRIBUTING.md](./CONTRIBUTING.md) for guidelines on how to contribute to the project. ### Citation: If you find this package useful, please cite as:

### Overview: OM1 is a modular AI runtime designed for developers to create and deploy multimodal AI agents across various digital and physical environments. ### Key Features: - **Modular Architecture**: Written in Go for performance and seamless integration. - **Data Input**: Easily handles new data and sensors. - **Hardware Support via Plugins**: Supports new hardware through plugins for API endpoints and specific robot hardware connections to `ROS2`, `Zenoh`, and `CycloneDDS`. - **Pre-configured Endpoints**: Supports Text-to-Speech, multiple LLMs from OpenAI, xAI, DeepSeek, Anthropic, Meta, Gemini, NearAI, Ollama (local), and multiple Visual Language Models (VLMs) with pre-configured endpoints for each service. - **Metrics & Observability**: Includes a pre-configured Prometheus and Grafana stack to monitor real-time AI pipeline metrics like LLM and ASR latencies. ### Tech Stack: Go ### Why It Matters: OM1 empowers developers to create highly capable human-focused robots, making it easy to upgrade and reconfigure them for different physical form factors. ### Repository: OpenMind/OM1 ### Primary Language: Go ### GitHub Stars: 2954 ### GitHub Description: Modular AI HAL (Hardware Abstraction Layer) for Robots ### Getting Started: Quick Start (5 Minutes) 1. **Install System Dependencies**: For macOS, install `brew` and `ffmpeg`. For Linux, install `portaudio19-dev` and `ffmpeg`. 2. **Download or Build