Projects

### Overview: 本书旨在为初学者提供自然语言处理（NLP）和深度学习的基础知识，涵盖这两个领域的重要主题。本书采用Python和PyTorch进行实现，旨在增强初学者的能力，特别是机器学习和深度学习的实践经验。 ### Key features: - **PyTorch 实现**：本书使用PyTorch进行自然语言处理和深度学习的实现。 - **代码示例**：每章中包含慷慨的端到端代码示例，邀请读者参与实践。 - **社区支持**：提供GitHub仓库地址、贡献指南、联系方式和下载链接。 ### Tech stack: Python, PyTorch ### Why it matters: 本书为初学者提供了自然语言处理和深度学习的基础知识，强调了实践的重要性，特别是机器学习和深度学习的实践经验。通过使用PyTorch，本书旨在增强初学者的能力，特别是机器学习和深度学习的实践经验。

### Overview: This repository contains hands-on, framework-free Colab notebooks designed specifically for the AI Engineer or Forward Deployed Engineer (FDE) skill set. The notebooks cover model APIs, structured output, tool calling, RAG, evals-as-the-spine, agents (loop from scratch, tool design, guardrails, MCP, Skills), fine-tuning vs LoRA, prompt-injection/security, LLMOps, and customer craft. The notebooks run on the free Groq API, providing a practical approach to transitioning from a backend or full-stack engineer to an AI Engineer, FDE, or Solutions Engineer. ### Key Features: - **Framework-free, on purpose**: You build the agent loop, RAG, and evals from raw API calls, understanding the underlying mechanics before using frameworks. - **Evals are the spine**: "Measure before you tune" is installed early, ensuring engineers can judge when to stop tuning and when to move on. - **Free to run, end to end**: Everything runs on the free Groq API, with optional Colab-GPU appendices verified on a real Colab T4. - **Real case studies, not toy demos**: Three end-to-end case studies showcase the skills under real constraints: a support assistant, a pipeline-vs-agent cost showdown, and a red-team robustness benchmark. - **OpenAI-compatible throughout**: Patterns transfer directly to OpenAI and Anthropic with small changes needed. ### Tech Stack: - Jupyter Notebook ### Why It Matters: These notebooks provide a practical, hands-on approach to the AI Engineer

### Overview: This project is a TensorFlow implementation of "Fast and Accurate Image Super Resolution by Deep CNN with Skip Connection and Network in Network," a deep learning-based Single-Image Super-Resolution (SISR) model. The model structure includes Deep CNN with Residual Net, Skip Connection, and Network in Network. It is designed to enhance image quality by utilizing parallelized 1x1 CNNs and depthwise separable convolutions. ### Key features: - Pixel Shuffler - Transposed-CNN (optional) - Self Ensemble - Clipping Normalization - Dynamically load training images - Add extra layers - Update default parameters for better PSNR result - Depthwise Separable Convolutions ### Tech stack: Python ### Why it matters: Our model, DCSCN, is significantly lighter than other Deep Learning-based SISR models, making it more efficient and faster. It achieves better image restoration performance, particularly in terms of PSNR, as demonstrated by the sample results provided.

Overview: This repository contains the official PyTorch implementation of "History-Guided Video Diffusion," presented at ICML 2025. The project introduces the Diffusion Forcing Transformer (DFoT) and History Guidance (HG) methods, enhancing video generation quality, temporal consistency, and motion dynamics. Key features: - Diffusion Forcing Transformer (DFoT) - History Guidance (HG) Tech stack: Python Why it matters: This project provides a comprehensive framework for generating videos conditioned on context frames, showcasing significant improvements in video quality and dynamics through the use of History Guidance.

### Overview: This guide serves as a comprehensive resource for developers looking to enhance their skills in machine learning and artificial intelligence. It covers essential topics such as machine learning frameworks, libraries, and tools, as well as practical applications and resources for learning and development. Whether you're a beginner or an experienced developer, this guide provides valuable insights and practical tips to help you master machine learning. ### Key Features: - **Machine Learning Frameworks**: Explore popular frameworks like TensorFlow, PyTorch, and others. - **Large Language Models (LLMs)**: Dive into training and deploying LLMs. - **Python**: Essential for machine learning development, with detailed guides and resources. - **LLMs Training Frameworks**: Learn about tools and libraries for training large language models. - **Tools for Deploying LLMs**: Discover the best tools for deploying and integrating LLMs into your projects. - **Running Large Language Models Locally**: Understand how to run LLMs locally for development and testing. ### Tech Stack: - Python ### Why It Matters: This guide is crucial for developers looking to leverage machine learning and artificial intelligence in their projects. It provides a solid foundation in machine learning concepts, frameworks, and practical applications, making it an invaluable resource for anyone interested in the field of machine learning and AI. ### Repository: mikeroyal/Machine-Learning-Guide - **Stars**: 711 - **Description**: A guide covering Machine Learning including the applications, libraries and tools that will make you better and more efficient with Machine Learning development. - **GitHub URL**:

Overview: VoxelHashing is a real-time 3D reconstruction algorithm designed for large-scale applications, showcasing significant advancements in large-scale, real-time 3D reconstruction at Siggraph Asia 2013. Key features: - High-performance voxel-based rendering - Real-time reconstruction capabilities - Scalability for large datasets Tech stack: C++ Why it matters: VoxelHashing is crucial for applications requiring high-speed and efficient 3D reconstruction, particularly in large-scale environments, making it a valuable tool in fields such as virtual reality, augmented reality, and large-scale data visualization.

Overview: QVAC is an open-source local AI SDK that enables developers to run a variety of AI tasks locally, including language models, speech-to-text, image and video generation, and more. It supports GGUF, RAG, image, music, and video generation, speech-to-text, P2P inference, and is compatible with Linux, macOS, Windows, Android, and iOS. Key features: - Supports GGUF, RAG, image, music, and video generation - Speech-to-text functionality - P2P inference capability - Cross-platform compatibility Tech stack: TypeScript Why it matters: QVAC simplifies the process of integrating AI into local applications by providing a comprehensive SDK that can run on any device, ensuring privacy and control over data. Its one-sdk approach makes it easy to integrate AI capabilities into existing projects and systems, making it a valuable tool for developers looking to build local AI solutions.

Overview: MaverickMCP is an open-source stock market MCP server designed for local analysis, connecting to Yahoo Finance data through `yfinance` with no API key required for core tools. It includes market data, technical analysis, portfolio tracking, and optional backtesting and research functionalities. Key features: - Market data and technical analysis - Stock screening - Portfolio tracking - Optional backtesting and research Tech stack: Python Why it matters: MaverickMCP provides a platform for educational and informational use, allowing users to analyze market data, track portfolios, and test strategies without placing orders or providing financial advice. Repository: wshobson/maverick-mcp GitHub stars: 710 GitHub description: Stock market MCP server for Yahoo Finance data, technical analysis, portfolio tracking, and Python backtesting. Runs locally with no API key for core tools.

### Overview: `pilotfish` is a multi-model orchestration policy designed for Claude Code, focusing on efficient and cost-effective execution of models while maintaining accountability and quality checks. ### Key features: - **Role Agents**: Handle volume work efficiently. - **Frontier Main Session**: Orchestrates and approves model execution. - **Verification Guards**: Ensure quality through fresh context reviewers. ### Tech stack: Python ### Why it matters: Most coding-session tokens are spent on search, repetitive edits, tests, and documentation rather than frontier judgment. `pilotfish` routes these bounded paths to cheaper roles while keeping the main session accountable and using fresh-context reviewers at material acceptance boundaries.

### Overview: Sketch Your Own GAN is a project that leverages off-the-shelf Generative Adversarial Networks (GANs) to customize images based on hand-drawn sketches. The method allows users to input one or a few hand-drawn sketches and modify an existing GAN model to match the input, preserving other visual cues such as color, texture, and background. ### Key Features: - **Customizable GAN Models**: The project utilizes pre-trained GAN models and customizes them to match input sketches. - **Interpolation and Editing**: Users can interpolate between different customized models or edit images using the customized models. - **Latent Space Interpolation**: The project demonstrates smooth interpolation in the latent space, allowing for seamless transitions between different customized models. ### Tech Stack: - **Python**: The project is primarily developed using Python, leveraging libraries such as PyTorch for deep learning. ### Why It Matters: - **User-Friendly Interface**: The project provides a user-friendly interface for anyone to customize GAN models based on their sketches, making it accessible to a wide range of users. - **Preservation of Visual Cues**: The method ensures that other visual elements such as color, texture, and background are preserved after modifications, making the customized images more realistic and visually appealing. ### Repository: - **Repository**: [PeterWang512/GANSketching](https://github.com/PeterWang512/GANSketching) - **GitHub Stars**: 710 - **GitHub Description**: Sketch Your Own GAN:

### Overview: This repository houses annotated datasets of relationships between entities/nominals in both English and Portuguese. It includes datasets from various sources, such as SemEval, Wikipedia, and BioNLP, aimed at training supervised models for semantic relationship extraction. ### Key Features: - **Traditional Information Extraction (TIE)**: Relationships are manually annotated and belong to pre-determined types. - **Open Information Extraction (OIE)**: Relationships are manually annotated but do not have specific types. - **Distantly Supervised (DS)**: Relationships are annotated using distant supervision techniques. ### Tech Stack: Python, TensorFlow, PyTorch, and other machine learning libraries. ### Why It Matters: This repository provides a rich collection of annotated datasets for researchers and developers working on semantic relationship extraction tasks. It serves as a valuable resource for training models and improving the accuracy of relation extraction in various languages and contexts.

Overview: SearchGPT is an open-source project that builds a grounded search engine based on LLM technology, providing natural language answers. It supports web search and file content search, and integrates with OpenAI embeddings. Key features: - Web search with real-time results - File content search (PPT/DOC/DOCX/PPTX, etc.) - Semantic search using OpenAI embeddings - LLM integration: OpenAI or GooseAI - Easy-to-use frontend interface Tech stack: Python Why it matters: This project is a minimal implementation of a new Bing search engine, focusing on providing accurate and grounded answers to user queries. Repository: MichaelThwan/searchGPT Primary language: Python GitHub stars: 711 GitHub description: Grounded search engine based on LLM / ChatGPT / OpenAI API. It supports web search, file content search etc. README excerpt: searchGPT - An open-source project to build a search engine based on Large Language Model (LLM) technology to give natural language answers. It supports using sources like the Internet and supports file content search. Please give me a star if you like it!

### Overview: CV_PaperDaily is a repository containing notes and references related to computer vision papers, updated irregularly. It includes a diverse range of topics such as object detection, instance segmentation, and more, with an emphasis on practical applications and challenges. ### Key Features: - Notes and references on various computer vision papers. - Updated regularly with the latest research in computer vision. - Focuses on practical applications and challenges in computer vision. ### Tech Stack: Unknown ### Why it Matters: This repository serves as a valuable resource for researchers and developers interested in computer vision, providing insights into cutting-edge research and practical applications in the field.

Overview: 语音识别和语音合成的REST API示例项目。 Key features: - 语音识别 REST API 示例 - 语音合成 REST API 示例 Tech stack: Java Why it matters: 本项目展示了如何使用百度语音识别和语音合成的REST API，适用于多种操作系统和编程语言，通过HTTP请求即可调用。 GitHub stars: 709

Overview: Magnetron is a zero-dependency machine learning framework built in C, offering a modern Python API for full control over execution and memory management. Key features: - Tensor system - Execution model - Operator backend - CPU backend (multi-dispatch design) - CUDA backend (in progress) - Serialization - Backend extensibility Tech stack: C Why it matters: Magnetron provides a compact, inspectable runtime with a clean base for experimentation and research, making it ideal for understanding and optimizing machine learning models.

### Overview: This repository is a curated list of resources focused on Machine Learning in Geospatial Data Science. It includes guides, papers, books, courses, and code projects that cover various aspects of applying deep learning and semantic segmentation techniques to geospatial data. ### Key Features: - **Code Projects and Workflows**: Includes projects and workflows for semantic segmentation, image classification, and object detection using deep learning. - **Datasets**: A collection of datasets for satellite imagery feature detection, including Dstl Satellite Imagery Feature Detection. - **Papers**: A list of research papers on deep learning and semantic segmentation in geospatial data science. - **Books**: Recommended books on machine learning and geospatial data science. - **Courses**: Information on courses related to geospatial data science and machine learning. ### Tech Stack: Python, TensorFlow, Keras, PyTorch, Numpy, Pandas, Matplotlib, Scikit-learn, OpenCV, and other geospatial data science tools. ### Why it Matters: This repository serves as a comprehensive resource for geospatial data scientists and researchers interested in applying machine learning techniques to geospatial data. It provides practical insights and tools for semantic segmentation, image classification, and object detection, making it a valuable tool for anyone looking to enhance their geospatial data science skills.

### Overview: QuietHarness 是一个轻量级的 AI 编程工作系统，旨在提供可靠性和可回滚的边界，帮助开发者在使用不同 AI 编程工具（如 Claude Code、Codex、Cursor）时保持代码质量和一致性。它通过压缩核心可靠性边界和提供干运行、备份、隔离测试和可逆安装等功能，简化了切换工具时的准备工作。 ### Key features: - **干运行验证**：在不修改现有代码的情况下，验证任务的正确性。 - **备份和隔离测试**：提供任务的备份和隔离测试功能，确保任务的完整性和可逆性。 - **可回滚安装**：安装器会保留旧的配置文件并提供回滚功能，确保系统的可恢复性。 - **项目事实管理**：任务事实和项目状态保持在项目文件中，不干扰日常开发流程。 - **业务和任务连续性**：业务和任务的连续性通过明确的事实源和写回协议来维护。 ### Tech stack: QuietHarness 使用 Python 编写，依赖于 Claude Code、Codex、Cursor 和其他 AI 工具的 API。 ### Why it matters: QuietHarness 提供了一种简单、可靠且可回滚的方式来管理 AI 编程工具的切换，帮助开发者在使用不同工具时保持代码质量和一致性。它简化了工具切换的准备工作，减少了对私人系统的依赖，并通过干运行验证和备份功能确保了任务的完整性和可逆性。

### Overview: LLMFlows is a framework designed for building simple, explicit, and transparent Large Language Model (LLM) applications. It simplifies the process of integrating LLMs and vector stores, ensuring that each component is easily monitored, maintained, and debugged. The framework provides a minimalistic set of abstractions that allow users to create well-structured and readable code without hidden prompts or LLM calls. ### Key Features: - **Minimalistic Abstractions**: Simplifies the integration of LLMs and vector stores. - **Explicit API**: Ensures clean, readable code with full control over LLM interactions. - **Transparent Flows**: Provides traceable flows and complete information for each app component. - **Simple and Explicit**: Focuses on building flexible LLM-powered apps without compromising on capabilities. - **Transparent**: Ensures full transparency on LLM-powered apps, making it easy to monitor, maintain, and debug. ### Tech Stack: - Python ### Why It Matters: - **Simplicity**: Builds a simple, well-documented framework with minimal abstractions. - **Explicitness**: Provides an explicit API for writing clean and readable code. - **Transparency**: Ensures complete transparency on each app component, making it easy to monitor, maintain, and debug. ### GitHub Stars: 708 ### Documentation: - [LLMFlows ReadTheDocs](https://llmflows.readthedocs.io) ### Installation: ```bash pip install llmflows ``` ### Philosophy: - **Simple**: Focuses on building a simple

### Overview: MovieChat is a cutting-edge method for long video understanding, achieving state-of-the-art performance across various benchmarks. It leverages sparse memory techniques to significantly reduce GPU memory usage, making it highly scalable for large videos. ### Key Features: - **Sparse Memory Representation**: MovieChat employs sparse memory to efficiently represent video features, achieving a 10000× advantage in GPU memory cost per frame compared to other methods. - **End-to-End Learning**: The model is trained end-to-end, enabling seamless integration with existing video understanding tasks. - **Scalability**: MovieChat is capable of handling videos with over 10K frames on a 24GB graphics card, demonstrating its robustness and efficiency. ### Tech Stack: - Python ### Why It Matters: MovieChat stands out by providing a significant improvement in GPU memory usage, making it highly scalable and efficient for large video processing tasks. Its end-to-end learning approach ensures seamless integration with existing video understanding frameworks, making it a valuable addition to the field of video analysis and understanding.

Overview: Active Graph is an event-sourced reactive graph runtime designed for durable and stateful agents. It enables behaviors to react to a shared graph instead of talking to each other, ensuring traceability and resumability of runs. Key features: - Event-sourced reactive graph runtime - Long-running, auditable, and stateful systems - Behaviors react to a shared graph instead of talking to each other - Every change is traceable - Every run is resumable, forkable, and diff-able from its event log Tech stack: Python Why it matters: Active Graph provides a unique approach to agent-based systems, focusing on event-sourcing and reactive graph processing, making it suitable for applications requiring high durability and traceability. Repository: yoheinakajima/activegraph Primary language: Python GitHub stars: 707 GitHub description: Event-sourced graph runtime for long-running, auditable, and stateful agents