Projects

### Overview: A comprehensive guide to Machine Learning for beginners ### Key features: - **Prerequisites**: - Basic mathematics: Linear Algebra, Matrix Algebra, Probability and Statistics, and Calculus. - Programming fundamentals: Data Structures and Algorithms. - **Tech stack**: The roadmap covers various tools and languages used in Machine Learning, including Python, TensorFlow, Keras, and Jupyter Notebooks. - **Why it matters**: This guide is essential for anyone looking to start their journey into Machine Learning, providing a structured path from basics to advanced concepts. ### Tech stack: Python, TensorFlow, Keras, Jupyter Notebooks ### Why it matters: This guide is essential for anyone looking to start their journey into Machine Learning, providing a structured path from basics to advanced concepts.

### Overview: Visionary is an open-source platform built on WebGPU and ONNX Runtime, enabling real-time rendering of diverse Gaussian Splatting variants (3DGS, MLP-based 3DGS, 4DGS, Neural Avatars, and future algorithms) and traditional 3D meshes directly in the browser. It supports dynamic neural processing and feedforward generative post-processing, making it highly efficient and user-friendly. ### Key Features: - **⚡️ Native WebGPU Powered**: Utilizes `webgpu` to achieve high-performance parallel sorting and rendering of millions of Gaussian particles. - **🎨 Hybrid Rendering Architecture**: Automatically handles depth mixing (Depth Compositing) between Gaussian point clouds and standard Meshes, perfectly solving occlusion issues and supporting complex scene compositions. - **📦 Universal Asset Loader**: Single interface to integrate seamlessly with existing web applications. ### Tech Stack: - Python ### Why It Matters: Visionary simplifies the process of rendering complex 3D scenes, making it easier to reproduce, compare, and deploy various 3DGS-family methods. It unifies inference and rendering directly in the browser, significantly lowering the barrier to entry for both reconstructive and generative paradigms.

### Overview: This repository provides a TensorFlow-based implementation of the paper "PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume," by Deqing Sun et al. (CVPR 2018). The implementation includes both TensorFlow-based training and inference, is portable across Linux and Windows, supports multi-GPU training, and allows for mixed-precision training. The code achieves better results on the challenging MPI-Sintel 'final' dataset compared to the official paper. ### Key features: - TensorFlow-based training and inference - Portable across Linux and Windows - Supports multi-GPU training - Mixed-precision training - Better results on MPI-Sintel 'final' dataset compared to the official paper ### Tech stack: TensorFlow ### Why it matters: This implementation provides a comprehensive and efficient solution for optical flow prediction, making it a valuable resource for researchers and practitioners in computer vision and related fields.

Overview: Poplar is a web-based annotation tool designed for natural language processing (NLP) that aims to provide a user-friendly interface for annotators. Key features: - Real-time collaboration - Automatic tagging - Customizable annotation tools Tech stack: TypeScript Why it matters: Poplar simplifies the process of annotating large volumes of text data, making it easier for researchers and professionals to extract insights from unstructured information. Repository: synyi/poplar Primary language: TypeScript GitHub stars: 530 GitHub description: A web-based annotation tool for natural language processing (NLP) README excerpt:

### Overview: This repository contains multiple implementations of abstractive text summarization models using Google Colab. The models are designed to summarize text in multiple languages, including Arabic, Hindi, and Amharic. The project aims to provide a comprehensive guide for beginners and advanced users on how to work with Google Colab and integrate it with Google Drive, making it easy to run the examples without the need for a powerful machine. ### Key Features: - **Implementation A**: Cornerstone seq2seq model with attention using Bidirectional LSTM Encoder and decoder. - **Implementation B**: Seq2seq model with Pointer Generator. - **Implementation C**: Seq2seq model with Reinforcement Learning. ### Tech Stack: - Python (Jupyter Notebook) ### Why It Matters: This project is valuable for developers and researchers interested in abstractive text summarization. It provides a practical guide on how to implement and use different models for summarization, making it easier to understand and apply the techniques in various languages. The repository is designed to be run on Google Colab, ensuring minimal setup and easy integration with Google Drive, making it accessible for users with varying levels of technical expertise.

Overview: The PS Move API is a cross-platform library that enables developers to track the PS Move Motion Controller using Bluetooth and USB, providing 3D space tracking and sensor fusion for augmented and virtual reality applications. Key features: - Pairing of Bluetooth controllers via USB - Setting LEDs and rumble via USB and Bluetooth - Reading inertial sensors and buttons via Bluetooth - Supporting extension devices (such as the Sharp Shooter and the Racing Wheel) - Tracking up to 5 controllers in 3D space via OpenCV - 3D orientation tracking via an open source AHRS algorithm - Sensor fusion for augmented and virtual reality applications Tech stack: C Why it matters: The PS Move API is essential for developers looking to create immersive and interactive experiences using the PS Move Motion Controller, offering advanced tracking and sensor fusion capabilities that enhance the gaming and entertainment industry. Repository: thp/psmoveapi GitHub stars: 530 GitHub description: Cross-platform library for 6DoF tracking of the PS Move Motion Controller. Sensor fusion, computer vision, ambient display (LED orb).

Overview: SwiftSpeech is a speech recognition framework designed specifically for SwiftUI, providing a simple and elegant way to integrate speech recognition functionality into iOS apps. Key features: - UI control + speech recognition functionality in just a few lines of code. - Customizable cancelling. - SwiftUI style reactive APIs and Combine support. - Highly customizable but also highly reusable via a composable structure. - Fully open low-level APIs. Tech stack: Swift, SwiftUI, Combine, Speech framework. Why it matters: SwiftSpeech simplifies the process of integrating speech recognition into iOS apps, making it easier to implement and more user-friendly.

### Overview: This repository contains the code and models for object detection and tracking from the CMU [DIVA](https://www.iarpa.gov/index.php/research-programs/diva) system. Our system achieves the best performance on the ActEv leaderboard, which is a benchmark for video action detection. The models are optimized for speed, using a Faster-RCNN backbone with a dilated CNN and FPN for object detection and a DeepSORT tracker for tracking. ### Key features: - **Object Detection**: Uses a Faster-RCNN model with a backbone of ResNet-101 and FPN for efficient and accurate detection. - **Tracking**: Implements a DeepSORT tracker for real-time tracking of objects. - **Multi-Camera Tracking and ReID**: Supports multi-camera tracking and ReID capabilities, including multi-thread and multiple-image batch inferencing for improved performance. ### Tech stack: Python ### Why it matters: This repository provides essential tools for researchers and developers working on surveillance video analysis, offering high performance and scalability, making it a valuable resource for applications in video surveillance, security, and autonomous systems.

### Overview: OmniSpace 是一个本地优先的全栈 AI 创作工作站，旨在提供从文本到图像、视频、语音和知识学习的完整创作管线。它结合了多种先进的技术，包括 vLLM、LLaMA.cpp 和 Transformers，以及一个自研的显存调度器和舱壁隔离系统，确保了高效和安全的推理。 ### Key features: - **本地推理**：所有模型推理都在本地 GPU 上进行，数据不出机房。 - **显存调度器**：自研显存协调器，支持 16GB 显卡的多条创作链路。 - **舱壁隔离与自愈**：每个重模块独立的健康账本，异常自动进入恢复阶梯。 - **数据全生命周期管理**：删除分级确认、存储清理中心、备份加密、操作审计、断电自恢复等。 - **高质量文档**：架构图、API 文档、部署手册、故障排查等。 - **插件式推理引擎**：支持 vLLM、LLaMA.cpp 和 Transformers 多后端热切换，以及 CuteMamen 插件系统。 ### Tech stack: Python ### Why it matters: OmniSpace 是一个完整的、工程化的、可落地的本地 AI 创作平台。它提供了一站式的创作工具，包括文本、图像、视频、语音和知识学习，旨在帮助创作者提升效率和质量。

Overview: 本书提供了一系列入门示例，旨在引导新手入门于基于LLM的AI代理开发，特别关注于当前LLM开发领域的快速进展。 Key features: - 介绍LLM基础 - 提供简单示例 - 引导新手开发AI代理 Tech stack: Jupyter Notebook Why it matters: 本书为初学者提供了实用的入门指南，帮助他们快速上手AI代理开发，是当前LLM开发领域的重要资源。 Repository: huangjia2019/ai-agents Primary language: Jupyter Notebook GitHub stars: 529 GitHub description: Introductory examples for building LLM-based AI agents.

Overview: PaL: Program-Aided Language Models is a project that introduces a new approach to solving reasoning problems using large language models (LLMs) and Python code generation. The key features include generating reasoning chains of text and code, offloading code execution to a Python interpreter, and implementing the approach with a few-shot prompting method. Key features: - Generating reasoning chains of text and code - Offloading code execution to a Python interpreter - Implementing the approach with a few-shot prompting method Tech stack: Python Why it matters: This project represents a novel method for leveraging LLMs to solve complex reasoning problems, particularly those involving arithmetic and procedural tasks, by generating text and code solutions. The use of Python as the backend provides a practical implementation that can be easily integrated with existing Python environments, making it accessible for a wide range of applications.

### Overview: Stylized-ImageNet is a dataset designed to induce a bias in Convolutional Neural Networks (CNNs) towards recognizing shapes rather than local textures. This dataset is based on the ImageNet dataset and has been used to improve the accuracy and robustness of CNNs by increasing their ability to recognize shapes. ### Key Features: - **Stylized Images**: Images are distorted to emphasize shapes over local textures. - **Dataset Creation**: The dataset is created by applying a stylization process to standard ImageNet images. - **Code Repository**: Contains information and code for creating and using the Stylized-ImageNet dataset. - **GitHub Stars**: 529, indicating the popularity and usefulness of the repository. ### Tech Stack: - Python ### Why It Matters: Stylized-ImageNet provides a useful dataset for researchers to test and improve the ability of CNNs to recognize shapes, rather than local textures. This can lead to more accurate and robust models, making it valuable for various applications in computer vision and image processing.

Overview: `vibe-rules` is a powerful CLI tool designed to manage and share AI rules across various editors and tools, enhancing workflow efficiency. Key features: - Save and load AI rules from local files or NPM packages. - Automatically install shared rules from project dependencies. - Apply rules to specific editors or project configurations. - Support for different editor formats like `cursor`, `claude-code`, `windsurf`, and `vscode`. Tech stack: TypeScript Why it matters: `vibe-rules` streamlines the process of managing AI rule configurations, making it easier to integrate and use AI prompts across different tools and environments, improving productivity and workflow efficiency.

Overview: NexusRAG is a hybrid RAG system that combines vector search, knowledge graph, and cross-encoder reranking into a seamless pipeline. It leverages Gemini or local Ollama models and supports Docling document parsing, image/table captioning, and inline citations. Key features: - **Docling / Marker Document Parsing**: Preserves headings, page boundaries, formulas, and layout. - **Image / Table Extraction**: Extracted and captioned by vision LLMs. - **Hybrid Semantic + Structural Chunking**: Preserves document structure. - **Dual-Model Embeddings**: BAAI/bge-m3 (1024d) + KG embedding (Gemini 3072d / Ollama / sentence-transformers). - **3-Way Parallel Retrieval**: Vector over-fetch, KG entity lookup, cross-encoder reranking. - **LightRAG Knowledge Graph**: Entity extraction, relationship mapping, multi-hop traversal. - **Auto-Generated Citations**: 4-char IDs with page number and heading path. - **End-to-End Page Awareness**: Preserved throughout the document. Tech stack: Python. Why it matters: NexusRAG provides a comprehensive solution for knowledge retrieval and document processing, offering advanced features like Docling parsing, image and table extraction, and structured citation generation, making it highly valuable for developers looking to enhance their applications with advanced NLP capabilities.

### Overview: This repository is a collection of awesome resources for individuals interested in data science. It provides a curated list of tools, books, and projects that can help developers and data enthusiasts explore and advance their skills in data science. ### Key Features: - Python and SQL resources - Data Science books and projects - Natural Language Processing (NLP) tools - Machine Learning and Deep Learning projects - Data Books and Computer Vision projects - Data Papers and Machine Learning Operations (MLOps) and Cloud projects ### Tech Stack: - Python - SQL ### Why It Matters: This repository serves as a valuable resource for anyone looking to deepen their understanding of data science. It provides a comprehensive collection of tools and projects that can be used to enhance skills in various aspects of data science, including Python, SQL, NLP, machine learning, and more. ### Repository: - **Repository:** multigroupco/DMG-Data-Science-Awesome - **Stars:** 528 - **Description:** Source for Who're Interested in Data! - **GitHub URL:** [multigroupco/DMG-Data-Science-Awesome](https://github.com/multigroupco/DMG-Data-Science-Awesome) - **GitHub Description:** Source for Who're Interested in Data! ### Links: - [Awesome](https://github.com/sindresorhus/awesome) - The Awesome project - [Made With Love](https://github.com/chetanraj/awesome-github-badges) - Made with love badge

### Overview: SCube: Instant Large-Scale Scene Reconstruction using VoxSplats SCube is a groundbreaking project that leverages the power of VoxSplats to instantly reconstruct large-scale scenes. This innovative approach allows for rapid and accurate scene reconstruction, making it an invaluable tool in various applications such as augmented reality, virtual environments, and more. ### Key Features: - **VoxSplats**: Utilizes voxel splats to efficiently reconstruct large scenes. - **Instant Reconstruction**: Achieves real-time performance with minimal latency. - **Versatile Applications**: Suitable for various use cases including augmented reality, virtual environments, and more. - **Community Contributions**: Includes contributions from leading researchers in the field. ### Tech Stack: Python SCube is built primarily using Python, leveraging the strengths of the language to ensure efficiency and scalability. ### Why It Matters: SCube's ability to instantly reconstruct large-scale scenes with minimal latency makes it a game-changer in the field of scene reconstruction. Its real-time performance and versatility make it a valuable tool for developers and researchers alike. ### Repository: nv-tlabs/SCube Repository: nv-tlabs/SCube Primary language: Python GitHub stars: 530 GitHub description: [NeurIPS 2024] SCube: Instant Large-Scale Scene Reconstruction using VoxSplats ### Installation To get started with SCube, follow these steps: 1. **Create a Folder**: ```bash mkdir SCube-release cd SCube-release git clone https://github.com/n

Overview: This project explores the application of deep learning techniques to predict Bitcoin price movements by analyzing historical tick data. The goal is to leverage artificial intelligence to enhance the accuracy of Bitcoin price predictions. Key features: - Downloading Bitcoin tick data - Converting data to 5-minute OHLC format - Training a simple AlexNet model on 20,000 samples with an accuracy of 70% for predicting price direction - Quantifying price movements and predicting them with a 99% accuracy - Training on a large dataset of 1,000,000+ samples - Integrating bar volumes into the generated OHLC image - Utilizing CNN attention to identify important features in the image Tech stack: The project utilizes Python, TensorFlow, and Caffe for implementing the deep learning models. Why it matters: This project demonstrates the potential of deep learning in financial market prediction, offering a novel approach to enhancing the accuracy of Bitcoin price forecasts.

### Overview: This repository is a collection of AI projects, providing a platform for developers to experiment with various AI techniques and algorithms. ### Key features: - 24 Puzzle - Adversarial Attacks on DNN - Email Spam Filter - Genetic Algorithm for Functional Maximum Problem - Million Queens - Neural Networks for Face Recognition ### Tech stack: Jupyter Notebook ### Why it matters: These projects showcase the practical applications of AI in various domains, making it a valuable resource for developers looking to enhance their skills in AI development.

### Overview: 大肥鱼是一款桌面化的DeepSeek Harness伴侣，专为Windows用户设计，提供实时Agent状态显示，帮助用户在各种应用中保持对任务的实时监控。它透明、无边框且始终置顶，确保在任何情况下都能在任务状态卡底部看到Agent的工作状态。 ### Key features: - **实时任务状态显示**：涵盖思考、查找、修改、执行、验证、等待确认、完成和错误等任务状态。 - **项目和步骤展示**：清晰展示项目和步骤，以及DHS提供的待办进度。 - **DeepSeek API余额显示**：在状态气泡和设置页显示DeepSeek的余额，区分API Key和已登录账户。 - **兼容性**：支持WebUI和官方桌面端的共享客户端，两个客户端使用各自的插件配置。 - **拖放操作**：支持拖放、摸头、戳身体、碰尾巴，以及大小、气泡、声音和减少动态设置。 - **DHS页面右下角显示**：可选在DSH页面右下角显示轻量角色，默认关闭。 ### Tech stack: JavaScript ### Why it matters: 大肥鱼通过透明、无边框的原生窗口设计，确保用户在编辑器、浏览器或其他应用中也能实时监控Agent的工作状态，提高了工作效率和用户体验。它的实时状态显示和透明设计使其成为DeepSeek Harness的完美伴侣，特别是在需要实时监控任务进度的场景中。

Overview: PromptInject is a framework designed to analyze the robustness of large language models (LLMs) against adversarial prompt attacks, specifically focusing on goal hijacking and prompt leaking. Key features: - A prosaic alignment framework for mask-based adversarial prompt composition - Examination of two types of attacks: goal hijacking and prompt leaking - Demonstrates the vulnerability of GPT-3 to simple adversarial inputs - Provides a quantitative analysis of LLMs' robustness to adversarial prompts Tech stack: Python Why it matters: PromptInject is a significant contribution to the field of LLM safety, providing valuable insights into the vulnerabilities of GPT-3 and other LLMs to adversarial prompt attacks.