Projects

Overview: Tiny CUDA Neural Networks is a self-contained framework designed for training and querying neural networks. It includes a highly efficient fully fused multi-layer perceptron and a versatile multiresolution hash encoding. Key features: - Fully fused multi-layer perceptron - Multiresolution hash encoding - Support for various input encodings, losses, and optimizers Tech stack: C++ Why it matters: Tiny CUDA Neural Networks offers a fast, efficient solution for training and querying neural networks, particularly for applications requiring high performance and accuracy.

### Overview: These tutorials are designed to provide practical, hands-on experience with PyTorch. They cover a variety of applications, including recurrent neural networks (RNNs) for natural language processing (NLP) and sequence-to-sequence models for machine translation. The tutorials are written in Jupyter Notebooks and include examples of modern techniques and data from the internet. ### Key Features: - **Recurrent Neural Networks (RNNs) for NLP**: Demonstrates classification, generation, and sequence-to-sequence translation using RNNs. - **Sequence-to-Sequence Models**: Explains how to use sequence-to-sequence networks for translation. - **Attention Models**: Covers attention mechanisms in neural machine translation. - **Other Applications**: Includes examples of other RNN applications such as sentiment analysis and timeseries data prediction. ### Tech Stack: PyTorch, a popular deep learning framework. ### Why It Matters: These tutorials are designed to be practical and relevant, providing a solid foundation in PyTorch through real-world applications. They are updated to ensure compatibility with the latest versions of PyTorch and are maintained by the official PyTorch tutorials repository. ### Repository: The repository is deprecated and no longer maintained. It can be found at [GitHub](https://github.com/pytorch/tutorials). ### Primary Language: The tutorials are written in Jupyter Notebooks. ### GitHub Stars: The repository has 4,538 stars on GitHub. ### Get Started: To run the tutorials on a fresh Linux or Mac machine, follow these steps: 1. Install Anaconda.

**Overview:** Tengine is a lightweight, high-performance inference engine designed specifically for embedded devices. It aims to provide efficient and fast deployment of deep learning models across various AIoT applications. **Key features:** - **C++ Core:** Utilizes C++ for core module development, optimized for embedded device resources. - **Modular Architecture:** Separates front-end and back-end for efficient CPU, GPU, and NPU deployment. - **Cross-Platform Support:** Supports multiple platforms, including Linux, Windows, and macOS. - **Model Repository:** Provides a model repository on Google Drive and a link to a Google Drive folder. - **Quantization Tools:** Offers tools for model quantization, including support for uint8 and int8 data types. - **AutoKernel Plugin:** Simplifies the process of deploying optimized operators into Tengine. - **Container Support:** Integrates with SuperEdge for easier management of edge computing environments. **Tech Stack:** C++ **Why it Matters:** Tengine simplifies the deployment of deep learning models on embedded devices, making it accessible and efficient for a wide range of applications.

### Overview: Accord.NET Framework is a comprehensive library for machine learning, statistics, artificial intelligence, computer vision, and image processing. It provides a unified API for .NET, supporting various algorithms and models, and is compatible with multiple platforms including Windows, Xamarin, Unity3D, and mobile applications. ### Key Features: - **Machine Learning Algorithms**: Support for various learning algorithms such as Support Vector Machines, Neural Networks, and more. - **Statistical Methods**: Includes statistical analysis and methods for data processing. - **Computer Vision**: Capable of image processing and computer vision tasks. - **Image Processing**: Provides tools for image manipulation and analysis. - **Extensibility**: The framework is designed to be extensible, allowing developers to easily integrate new algorithms and models. ### Tech Stack: .NET Framework ### Why It Matters: Accord.NET Framework is a valuable resource for developers looking to implement machine learning, statistics, and computer vision algorithms in their .NET projects. Its unified API and extensibility make it a versatile tool for various applications, including scientific computing, image processing, and computer vision tasks.

### Overview: CISO Assistant is a comprehensive GRC platform designed to simplify and modernize cybersecurity management practices. It supports 200+ global frameworks, including ISO 27001, NIST CSF, SOC 2, CIS, PCI DSS, NIS2, DORA, GDPR, HIPAA, CMMC, and more. ### Key Features: - **Centralized Hub**: Connects multiple cybersecurity concepts with smart linking between objects. - **Multi-paradigm Tool**: Adapts to different backgrounds, methodologies, and expectations. - **Decoupled Compliance**: Separates compliance from cybersecurity controls, enabling reusability. - **Reusability and Interlinking**: Promotes reusability and interlinking instead of redundant work. - **API-first Approach**: Supports both UI interaction and external automation. - **Built-in Standards and Controls**: Comes packed with a wide range of built-in standards, security controls, and threat libraries. - **Customizable Frameworks**: Offers an open format to customize and reuse your own objects and frameworks. - **Risk Assessment and Remediation**: Includes built-in risk assessment and remediation tracking workflows. - **Custom Frameworks**: Supports custom frameworks via a simple syntax and flexible tooling. - **Rich Import/Export Capabilities**: Supports rich import/export capabilities across various channels and formats (UI, CLI, Kafka, reports, etc.). ### Tech Stack: Python ### Why It Matters: CISO Assistant aims to create a modern, integrated solution for cybersecurity management, addressing the fragmentation and redundancy issues faced by

### Overview: Embabel is a framework for authoring agentic flows on the JVM that seamlessly integrates LLM-prompted interactions with code and domain models. It supports intelligent path finding towards goals and is written in Kotlin, offering a natural Java-like usage model. ### Key Features: - **Intelligent Path Finding**: Embabel helps in finding the most efficient paths towards goals. - **LLM-Prompted Interactions**: It integrates LLM-prompted interactions seamlessly with the framework. - **Code and Domain Models**: Supports code and domain models for a comprehensive approach. ### Tech Stack: - **Kotlin**: Written in Kotlin. - **Java**: Offers a natural Java-like usage model. - **Spring**: Supports the framework. - **Spring Boot**: Framework for building and deploying applications. - **Apache Tomcat**: Web server for hosting applications. - **Apache Maven**: Dependency management for the project. - **JUnit**: Testing framework for verifying the correctness of the framework. - **ChatGPT**: For generating documentation and explanations. - **Jinja**: Template engine for rendering dynamic content. - **JSON**: For data interchange. - **GitHub Actions**: Continuous integration and deployment. - **SonarQube**: Quality gate for code analysis. - **Docker**: Containerization for deployment. - **IntelliJ IDEA**: IDE for development. ### Why It Matters: Embabel is designed to be a powerful tool for developers looking to integrate artificial intelligence and machine learning into their Java applications. It simplifies the process of integrating language models into their code, making it easier to

Overview: This project aims to develop a method for large pose 3D face reconstruction from a single image using Direct Volumetric CNN Regression. Key features: - Direct Volumetric CNN Regression - Large Pose 3D Face Reconstruction - Single Image Input Tech stack: MATLAB Why it matters: This project offers a novel approach to 3D face reconstruction, potentially revolutionizing the field by enabling more accurate and efficient reconstruction from single images, which is crucial for applications like augmented reality, medical imaging, and security systems.

### Overview: A simplified, highly flexible, commented, and easy-to-understand implementation of self-play based reinforcement learning based on the AlphaGo Zero paper. It is designed to be adopted for any two-player turn-based adversarial game and any deep learning framework of your choice. The implementation includes sample implementations for Othello, GoBang, TicTacToe, and more. ### Key Features: - Sample implementation for Othello in PyTorch and Keras - Easy-to-adopt framework for any game and framework - Sample implementation for GoBang and TicTacToe - Sample implementation for Dots and Boxes - Sample implementation for RTS game - Sample implementation for 3D TicTacToe - Sample implementation for 6x6 Othello - Sample implementation for 4x4 Othello - Sample implementation for 5x5 Othello - Sample implementation for 8x8 Othello - Sample implementation for 10x10 Othello - Sample implementation for 12x12 Othello - Sample implementation for 15x15 Othello - Sample implementation for 20x20 Othello - Sample implementation for 25x25 Othello - Sample implementation for 30x30 Othello - Sample implementation for 40x40 Othello - Sample implementation for 50x50 Othello - Sample implementation for 60x60 Othello -

### Overview: Kiro Crew is a persistent workspace designed for developers, running locally or remotely on your hardware. It remembers work across sessions, runs unattended, and integrates with various tools like Slack and Discord. ### Key features: - Persistent workspaces - Self-learning and self-evolving - Runs on local or remote hardware - Self-improving and continues beyond one session - Supports multi-step tasks and recurring jobs - Remembers work across sessions - Integrates with Slack, Discord, and other tools - Runs unattended, recurring jobs, and heartbeats ### Tech stack: Python ### Why it matters: Kiro Crew is a powerful development workspace that ensures your work continues even when you're not actively using it. It's designed to be persistent and self-improving, making it a valuable tool for developers who need reliable, uninterrupted work environments.

### Overview: M-flow is a bio-inspired cognitive memory engine that operates as a new paradigm for Graph RAG. It structures context through a graph that scores evidence paths, rather than relying on similarity or structure. ### Key features: - **Graph Topology Encodes Relevance**: The graph encodes the relevance of the graph structure rather than relying on similarity or structure. - **Relevance Is Not a Score**: M-flow scores evidence paths, not a similarity score, which is more aligned with the concept of relevance. - **Cognitive Memory System**: M-flow operates similarly to a cognitive memory system, where the graph takes over to propagate evidence along typed, semantically weighted edges. ### Tech Stack: Python ### Why it Matters: M-flow provides a new paradigm for Graph RAG by structuring context through a graph that scores evidence paths, rather than relying on similarity or structure. This approach ensures that the system can connect the query to the answer through a coherent structure of evidence, making it more relevant and useful for real-world applications.

### Overview: Unity-MCP is a comprehensive suite of tools and scripts designed for Unity game development. It includes AI skills, MCP tools, and a command-line interface for efficient and streamlined development. ### Key Features: - **AI Skills**: Enhances game AI with advanced techniques and algorithms. - **MCP Tools**: Simplifies complex Unity scripts into tools for faster development. - **CLI Setup**: Quick and easy setup using the CLI for a seamless development environment. ### Tech Stack: - **C#**: The primary programming language used for development. - **Unity Engine**: Works seamlessly with Unity for Unity game development. ### Why It Matters: Unity-MCP provides a complete toolkit for game developers, making AI development and game development more efficient and accessible. The CLI setup allows for quick and easy integration, and the AI skills and MCP tools streamline the development process, making it easier to create and test games.

### Overview: Monodepth2 is a PyTorch implementation for training and testing depth estimation models using the method described in "Digging into Self-Supervised Monocular Depth Prediction". This code is designed for non-commercial use and provides a reference implementation for the research presented in the paper. ### Key features: - **Model Training**: Monodepth2 supports training depth estimation models using a single image, stereo images, or both. - **Model Resolutions**: Available resolutions include 640x192, 1024x320, and combinations for stereo images. - **Evaluation Metrics**: The model achieves low absolute and relative errors in depth estimation, with a KITTI evaluation metric of 0.115 for mono and 0.109 for stereo. ### Tech stack: PyTorch 0.4.1, CUDA 9.1, Python 3.6.6, and Ubuntu 18.04. ### Why it matters: Monodepth2 offers a practical solution for depth estimation tasks, particularly useful in applications requiring accurate depth information from single or stereo images. The method presented in the paper is designed to be self-supervised, making it a valuable tool for researchers and developers looking to improve depth estimation capabilities in their projects.

### Overview: SimCLR is a simple yet effective framework for contrastive learning of visual representations. It has been shown to be strong semi-supervised learners, achieving state-of-the-art performance on various benchmarks. ### Key features: - **Self-Supervised Learning**: SimCLR uses contrastive loss to learn representations without labeled data. - **Pre-trained Models**: The repository includes 65 pretrained models for SimCLRv2, corresponding to the models described in the paper "SimCLRv2". - **Fine-Tuning**: Models can be fine-tuned on different levels of labeled data (1%, 10%, 100%) to improve their performance. - **Google Cloud Storage**: Pretrained and fine-tuned models are stored in Google Cloud Storage for easy access and deployment. ### Tech stack: Python and TensorFlow ### Why it matters: SimCLR offers a simple yet powerful approach to semi-supervised learning, making it a valuable tool for researchers and practitioners looking to enhance their models with minimal labeled data.

Overview: 金销数据云智能客服是一款专业的电商SaaS智能客服平台，专为电商行业量身定制。它支持微信、千牛、抖音、飞鸽、拼多多、小红书、京东、淘宝、京麦等多平台统一管理，提供智能问答、预设回复、多媒体支持、知识库定制和插件生态等功能，助力业务运营和决策优化。 Key features: - 支持微信、千牛、哔哩哔哩、抖音企业号、抖音、抖店、拼多多、微博聊天、小红书专业号运营、小红书、知乎等主流电商平台 - 基于AI大模型，智能理解客户咨询，自动生成专业回复 - 支持自定义常见问题回复模板，提升客服响应效率 - 支持文本、语音、图片等多种消息类型的智能处理 - 支持上传专属知识库文件，打造个性化智能客服机器人 - 支持丰富的插件系统，访问外部资源和自定义功能扩展 - 支持基于自有知识库定制企业 AI 应用 Tech stack: TypeScript Why it matters: 金销数据云智能客服平台通过深度理解电商业务需求，提供针对性解决方案，支持多平台统一管理，智能问答、预设回复、多媒体处理等功能，助力电商运营和业务决策优化。同时，平台支持多种大语言模型，灵活定制，满足不同业务场景需求。

Overview: Judge0 is an open-source online code execution system designed for humans and AI, enabling robust, fast, scalable, and sandboxed execution of AI-generated code. Key features: - Self-hostable or fully managed (SaaS) deployment - Quick and easy installation - Rich and verbose API documentation - Simple HTTP JSON API for easy integration - Official Python SDK for easy integration - Scalable architecture for handling high loads - Sandboxed compilation and execution of untrusted code - Support for 90+ languages - Compilation and execution of multi-file programs - Support for additional files alongside the single-file user's program - Support for custom user-defined compiler options, command-line arguments, and time and memory limits - Detailed execution results - Webhooks (HTTP callbacks) Tech stack: HTML Why it matters: Judge0 provides a robust, fast, scalable, and sandboxed online code execution system for humans and AI, making it ideal for various applications such as AI agents, competitive programming platforms, e-learning platforms, and more. Repository: judge0/judge0 GitHub stars: 4490 GitHub description: Robust, fast, scalable, and sandboxed open-source online code execution system for humans and AI.

Overview: This repository is dedicated to providing simple and ready-to-use tutorials for TensorFlow, aiming to make the learning process accessible and efficient. Key features: - Source code - Documentation - Wiki Tech stack: Python Why it matters: TensorFlow is a powerful deep learning framework, and this repository aims to simplify its usage, making it more accessible to developers and beginners alike.

Overview: This open-source, self-hosted AI agent automates the entire process of creating and publishing a YouTube channel, from researching topics to generating visuals and publishing after approval. Key features: - Automates YouTube channel creation and management - Generates scripts, narration, and visuals - Optimizes SEO and publishes videos after approval - Supports multiple text and AI video providers - Learns from analytics and user feedback Tech stack: JavaScript Why it matters: This agent provides a seamless, self-hosted solution for creating and managing YouTube channels, ensuring compliance with media rights and user approval, and continuously improving based on performance metrics and audience feedback. Repository: darkzOGx/youtube-automation-agent GitHub stars: 4487 GitHub description: 🎬 Open-source, self-hosted AI agent that runs a YouTube channel end to end: researches topics, writes scripts, generates narration, visuals and video, optimizes SEO, and publishes after your approval, then learns from analytics. Works with Gemini, OpenAI, OpenRouter, DeepSeek and more.

Overview: Merlion is a Python library designed for time series intelligence, providing a comprehensive framework for various time series learning tasks such as forecasting, anomaly detection, and change point detection. Key features: - Standardized and easily extensible data loading & benchmarking for a wide range of forecasting and anomaly detection datasets. - A library of diverse models for anomaly detection, forecasting, and change point detection, all unified under a shared interface. - Abstract `DefaultDetector` and `DefaultForecaster` models for efficient, robust performance. - AutoML for automated hyperparameter tuning and model selection. - Unified API for using a wide range of models to forecast with exogenous regressors. - Practical, industry-inspired post-processing rules for anomaly detectors. - Easy-to-use ensembles that combine the outputs of multiple models. - Flexible evaluation pipelines for both forecasting and anomaly detection. - Native support for visualizing model predictions. - Distributed computation backend using PySpark for industrial-scale applications. Tech stack: Python Why it matters: Merlion offers a one-stop solution for rapidly developing models for time series intelligence, providing engineers and researchers with a standardized and efficient framework for various time series learning tasks.

### Overview: DeepKE is an open-source toolkit designed for knowledge graph construction, specifically tailored for Chinese knowledge graphs. It supports entity, relation, and attribute extraction in various scenarios, including Chinese Schema, low-resource datasets, document-level processing, and multimodal integration. ### Key Features: - **Entity Extraction**: Extracts entities from text documents. - **Relation Extraction**: Identifies relationships between entities. - **Attribute Extraction**: Extracts attributes associated with entities. - **Supports Multiple Scenarios**: Suitable for Chinese knowledge graphs, low-resource datasets, document-level processing, and multimodal integration. ### Tech Stack: Python ### Why It Matters: DeepKE provides a comprehensive solution for knowledge graph construction, making it an essential tool for researchers and practitioners in natural language processing and knowledge graph engineering. Its flexibility and robustness make it suitable for a wide range of applications, from academic research to practical applications in information retrieval and knowledge discovery. ### Repository: Repository: zjunlp/DeepKE GitHub Stars: 4486 GitHub Description: [EMNLP 2022] An Open Toolkit for Knowledge Graph Extraction and Construction ### Primary Language: Python ### Table of Contents: - **Table of Contents** - **What's New** - **Prediction Demo** - **Model Framework** - **Quick Start** - **DeepKE-LLM** - **DeepKE-MCP-Tools** - **DeepKE** - **🔧 Manual Environment Configuration** - **🐳 Building With Docker Images** - **Requirements** - **Deep

### Overview: LMOps is a research initiative focused on enabling AI capabilities using large language models (LLMs) and generative AI models. It aims to improve prompt optimization, enhance context length, align LLMs, accelerate inference, and customize models for various domains. ### Key Features: - **Automatic Prompt Optimization (Promptist)**: Optimizes user input into model-preferred prompts using reinforcement learning. - **Structured Prompting**: Efficiently consumes long-sequence prompts by pre-appending retrieved documents or scaling in-context learning. - **LLM Alignment**: Aligns LLMs to improve performance and adapt them to specific domains. - **LLM Accelerator (Lossless Inference)**: Accelerates LLM inference without loss, improving performance. - **LLM Customization**: Adapts LLMs to domains for better performance. ### Tech Stack: Python ### Why It Matters: LMOps aims to enhance AI capabilities by leveraging large language models and generative AI, making it easier to develop AI products with foundational models. The initiative focuses on improving prompt optimization, context length, LLM alignment, inference speed, and domain-specific customization, ultimately leading to more effective and efficient AI applications.