Projects

### Overview: Vision Pilot is a free and fully open-source L2 ADAS stack powered by End-to-End AI technology. It is designed to be integrated with automotive OEMs and Tier-1 suppliers in series production passenger cars, and can be adopted for transportation and logistics use-cases in buses and trucks. ### Key Features: - **ACC (Autonomous Cruise Control)**: Autonomous cruise control. - **FCW (Forward Collision Warning)**: Forward collision warning. - **AEB (Autonomous Emergency Braking)**: Autonomous emergency braking. - **LKAS (Lane Keep Assist)**: Lane keep assist. - **LDW (Lane Departure Warning)**: Lane departure warning. - **ISA (Intelligent Speed Assist)**: Intelligent speed assist. - **Autopilot** (single-lane hands-free highway autopilot): Single-lane hands-free highway autopilot. ### Tech Stack: C++, PyTorch, OpenCV, ROS ### Why it Matters: Vision Pilot is a productionizable and safety-certifiable implementation of an open-source L2 ADAS system, designed to be integrated with automotive OEMs and suppliers in series production passenger cars, and can be adopted for transportation and logistics use-cases in buses and trucks. It is available under the Apache 2.0 license and can be freely used for both commercial and research purposes.

### Overview: Muse is a powerful text-to-image generator implemented using Pytorch. This project provides a comprehensive guide for training and using the Muse model, which leverages masked generative transformers to generate high-quality images from textual prompts. The code includes detailed instructions on setting up and training the VQGanVAE and MaskGit, making it accessible for developers looking to integrate this technology into their projects. ### Key Features: - **Muse Model Implementation**: The project details the implementation of Muse, a text-to-image generator using masked generative transformers. - **Pytorch Integration**: The code is written entirely in Pytorch, ensuring compatibility and ease of integration. - **Training Instructions**: Detailed steps on training a VQGanVAE and using it with a MaskGit, including hyperparameters and data handling. - **Image Generation**: Instructions on generating images from textual prompts using the MaskGit model. - **Super-Resolution Support**: The project includes options for super-resolution maskgit, allowing for higher-resolution images. ### Tech Stack: - **Python**: The primary language used for the project. - **Pytorch**: The deep learning framework used for training and generating images. ### Why It Matters: Muse offers a significant advancement in text-to-image generation by leveraging masked transformers, providing high-quality results that are both realistic and detailed. This technology has the potential to revolutionize various applications, from creative design to content creation, making it an exciting development in the field of artificial intelligence. ### Repository: The repository is accessible at <https://github.com/lucidrains/m

Overview: Lumina Note is a modern Markdown note-taking app that integrates a powerful AI assistant, enabling users to edit their notes directly in Markdown while seamlessly integrating AI for real work. Key features: - Live preview and reading mode - AI-powered agent for real work - WikiLinks, backlinks, and graph views - Flexibility in model routing Tech stack: TypeScript Why it matters: Lumina Note combines human-editable notes with AI assistance, making it a versatile tool for both personal and professional note-taking, while also providing a flexible platform for extending the app's functionality.

### Overview: NornicDB is a distributed low-latency, graph+vector, temporal multi-versioned consistent key-value store (MVCC) designed for AI and machine learning applications. It leverages Neo4j Bolt/Cypher and Qdrant's gRPC for seamless integration and intelligent features like schemas, managed embeddings, reranking, and GPU acceleration. ### Key Features: - **Graph Traversal**: Efficiently handles complex graph queries. - **Vector Retrieval**: Supports vector similarity searches. - **Temporal MVCC**: Maintains historical data with sub-millisecond consistency. - **Hybrid Graph+Vector**: Combines graph traversal with vector retrieval for comprehensive data analysis. - **GPU Acceleration**: Utilizes NVIDIA CUDA for enhanced performance on GPU hosts. - **Managed Embeddings**: Enables managed and optimized embeddings for efficient storage and retrieval. - **Reranking**: Improves search results through advanced ranking techniques. - **Policy-Based Memory Decay**: Automatically manages memory usage based on usage patterns. - **Multi-Arch Support**: Compatible with CPUs, CUDA, Metal, and Vulkan. ### Tech Stack: Go ### Why It Matters: NornicDB is designed to provide a unified database solution for AI applications that require graph, vector, and temporal data. It offers intelligent features like schema management, managed embeddings, and GPU acceleration, making it a powerful tool for modern AI workflows. The ability to seamlessly integrate with Neo4j and Qdrant makes it a versatile choice for developers looking to build scalable and intelligent systems.

Overview: PPT Agent 是一款基于软件工程理念的演示文稿全自动生成框架，通过严格的状态机驱动多 Agent 协作，从根源解决传统大模型生成的幻觉、重叠与布局混乱问题。 Key features: - 基于软件工程理念的演示文稿全自动生成框架 - 状态机驱动多 Agent 协作 - 从根源解决传统大模型生成的幻觉、重叠与布局混乱问题 - 无状态断点恢复 - 像构建软件工程一样生成演示文稿 Tech stack: Python Why it matters: PPT Agent 通过严格的状态机驱动多 Agent 协作，将演示文稿生成过程中的每一个环节都进行了系统化、模块化处理，从而解决了传统演示文稿生成过程中常见的幻觉、重叠和布局混乱等问题，使生成的演示文稿更加专业和规范。 GitHub stars: 916

Overview: A local memory engine for AI agents that captures and remembers conversations, providing a personal map of your interactions and context. Key features: - Captures every conversation word-for-word - Offers real-time recall and learning capabilities - Works with various AI agents like Claude, Code, Codex, OpenClaw, and more - Supports multiple languages and platforms (Python 3.11/3.12, macOS/Linux, Windows, MCP-compatible) Tech stack: Python, Rust, TypeScript, and the MCP protocol Why it matters: It simplifies your AI workflow by eliminating the need to remember previous conversations, making interactions more efficient and contextually relevant.

### Overview: Auto Browser is an MCP-native browser control plane designed for authorized workflows. It provides a shared Playwright browser with human takeover, reusable auth profiles, approvals, audit trails, and local-first deployment. This allows for a seamless and secure interaction with MCP clients, LLM agents, and operators. ### Key Features: - **Human Takeover**: Ensures the session remains intact when web content becomes brittle. - **Login Once, Reuse Later**: Saves named auth profiles to be reused across sessions. - **Local-First Deployment**: Runs the full stack on your own box with Docker Compose or Codespaces for a quick hosted demo. - **Safety Rails Built In**: Includes approvals, operator identity, PII scrubbing, witness receipts, and policy presets. - **Evidence Verification**: Witness receipts are Ed25519-signed and verified with `verify_witness_bundle.py`. - **Governed Skill Induction**: Verified browser traces can become staged skill candidates with provenance signed when a mesh identity is configured. ### Tech Stack: Python ### Why It Matters: - **MCP-native from Day One**: The browser surface is packaged as an MCP server, ensuring compatibility and seamless integration. - **Human Takeover**: Ensures the session remains intact when web content becomes brittle. - **Login Once, Reuse Later**: Saves named auth profiles to be reused across sessions. - **Local-First Deployment**: Runs the full stack on your own box with Docker Compose or Codespaces for a quick hosted demo. - **Safety Rails Built In**: Includes approvals,

Overview: RAG Time is a comprehensive 5-week learning journey designed to help developers master Retrieval-Augmented Generation (RAG). This project includes step-by-step guides, live coding samples, and expert insights, providing a complete package for beginners to advanced practitioners. Key features: - Step-by-step guides - Live coding samples - Expert insights - Ready-to-use implementations - Best practices - Engaging visuals Tech stack: Jupyter Notebook Why it matters: This project is essential for anyone looking to enhance their AI skills, stay ahead with cutting-edge retrieval techniques, and build smarter AI applications with RAG. The project includes interviews with AI leaders, tech talks, and practical demonstrations, making it a valuable resource for developers. Repository: microsoft/rag-time GitHub stars: 912 GitHub description: RAG Time: A 5-week Learning Journey to Mastering RAG

Overview: spaCy Layout is a plugin that integrates spaCy with Docling, enabling the processing of PDFs, Word documents, and other input formats into structured data. This allows for easy application of NLP techniques and structured data extraction. Key features: - Integrates with Docling to process PDFs and Word documents - Outputs structured data in a text-based format - Creates spaCy's familiar `Doc` objects for accessing labeled text spans and tables Tech stack: Python Why it matters: spaCy Layout simplifies the process of applying NLP techniques to structured data, making it easier to implement and scale. It's particularly useful for creating RAG pipelines and deep dives into converting PDFs and similar documents into structured data.

Overview: Jevos is an open-source alternative to Jev and Microsoft Decision 1, providing Yes/No decisions, choices, and scores with probabilities on your laptop's CPU. Key features: - Yes/No decisions - Multiple choice - Scores with probabilities Tech stack: C++ Why it matters: Jevos offers a lightweight, efficient solution for quick decision-making tasks, making it ideal for use on laptops or embedded systems where performance is critical. Repository: feder-cr/jev GitHub stars: 1357 GitHub description: An open-source alternative to Jev and Microsoft Decision 1, providing Yes/No decisions, choices, and scores with probabilities on your laptop's CPU. README excerpt: <div align="center"> <picture> <source media="(max-width: 374px) and (prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/feder-cr/jev/main/assets/banner-small-dark.gif"> <source media="(max-width: 374px)" srcset="https://raw.githubusercontent.com/feder-cr/jev/main/assets/banner-small-light.gif"> <source media="(max-width: 1239px) and (prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/feder-cr/jev/main/assets/banner-phone-dark.gif"> <source media="(max-width: 1239px)" srcset="https://raw.githubusercontent.com/feder-cr/jev/main/assets/banner-phone-light.gif"> <source

### Overview: The Berkeley Neural Parser is a high-accuracy natural language processing (NLP) parser designed for 11 languages. It leverages models from the 2018 ACL conference and includes updated pre-trained models and improved inference capabilities. ### Key Features: - **11 Languages Support**: Supports English, German, Korean, Polish, Swedish, Dutch, Portuguese, Spanish, French, Italian, and Japanese. - **Python Implementation**: Written in Python, making it easy to integrate with other Python projects. - **Updated Models**: Version 0.2.0 introduces higher-quality pre-trained models and improved inference using PyTorch. - **SpaCy Integration**: Recommended for use with spaCy, allowing seamless integration with NLP pipelines. - **Training Models**: Includes detailed instructions on training and using your own parsing models. ### Tech Stack: - Python - PyTorch ### Why It Matters: The Berkeley Neural Parser offers advanced parsing capabilities for multiple languages, making it a valuable tool for researchers and developers working on NLP tasks. Its updated models and improved inference methods ensure better accuracy and efficiency in parsing tasks.

### Overview: This project showcases multiple real-time diffusion models pipelines utilizing Diffusers, specifically focusing on the Latent Consistency Model (LCM) and its integration with LoRA (Low-Rank Adaptation). The demo includes live demonstrations and showcases various pipelines for image-to-image and text-to-image transformations, with the ability to use ControlNet for enhanced control and LoRA for faster inference times. ### Key Features: - **Real-Time Latent Consistency Model (LCM)**: Demonstrates the use of LCM with Diffusers, enabling real-time image-to-image and text-to-image transformations. - **ControlNet Integration**: Utilizes ControlNet for enhanced control over the generated images. - **LoRA Integration**: Implements LoRA to speed up inference times, particularly for SDXL and SDXL Turbo models. - **Pipeline Options**: Offers multiple pipelines including LCM, SD15, SDXL Turbo, and SDTurbo, each with specific configurations for different models. - **Environment Variables**: Allows setting various environment variables for customization and optimization. ### Tech Stack: - **Python**: The primary programming language used for the project. - **Docker**: Used for running the server locally. ### Why It Matters: This project is significant for developers interested in real-time image generation and control, offering a practical demonstration of cutting-edge AI techniques like LCM, LoRA, and ControlNet. It showcases the integration of these technologies with popular models like SD15, SDXL, and SDTurbo, making it a valuable resource for developers looking to implement similar functionalities in their projects.

Overview: This project reviews and analyzes the state-of-the-art methods and challenges in change detection based on artificial intelligence, focusing on applications in urban planning, environmental monitoring, agriculture investigation, disaster assessment, and map revision. The review includes various sensors, open datasets, and commonly used networks for AI-based change detection methods. Key features: - Implementation process of AI-based change detection - Available codes for AI-based methods - Data from different sensors for change detection Tech stack: Python, PyTorch, TensorFlow, Keras, Caffe Why it matters: This review helps researchers understand the field of AI-based change detection, including its state-of-the-art methods, applications, and challenges.

### Overview: iowncode is a curated collection of iOS, ML, AR resources, and UI additions. It includes SwiftUI, CoreML, CreateML, Vision Framework, and more. The repository is primarily written in Swift and has been starred by 912 developers. ### Key Features: - **SwiftUI**: Resources for building SwiftUI apps, including charts, gestures, progress bars, and alternative icons. - **CoreML and CreateML**: Techniques for training and deploying models on iOS devices, including real-time style transfer and NSFW image classifiers. - **Vision Framework**: Features for image recognition, similarity, and credit card scanning. - **ARKit**: AR resources for augmented reality applications. ### Tech Stack: Swift ### Why it Matters: This repository provides a comprehensive set of resources for iOS developers, covering everything from SwiftUI and CoreML to Vision Framework and ARKit. It's ideal for developers looking to enhance their skills in iOS development, machine learning, and augmented reality.

Overview: Texo is a minimalist, free, and open-source SOTA LaTeX OCR model that only requires 20M parameters. It is designed to be fast and lightweight, making it accessible and precise for note-taking. Key features: - Free and open-source - Fast and lightweight inference - Trainable on consumer's-level GPU - Well-organized code as a tutorial - Runs in browser Tech stack: Python Why it matters: Texo is a comprehensive practice for combining knowledge and experiences, offering a free and accessible solution for LaTeX OCR, which is essential for STEM and AI learners. Repository: alephpi/Texo Primary language: Python GitHub stars: 910 GitHub description: A minimalist free and open-source SOTA LaTeX OCR model which contains only 20M parameters. It is designed to be fast and lightweight, making it accessible and precise for note-taking. The model is trainable on consumer's-level GPU and runs in browser. It is a comprehensive practice for combining knowledge and experiences, offering a free and accessible solution for LaTeX OCR, which is essential for STEM and AI learners. Repository: alephpi/Texo Primary language: Python GitHub stars: 910 GitHub description: A minimalist free and open-source SOTA LaTeX OCR model which contains only 20M parameters. It is designed to be fast and lightweight, making it accessible and precise for note-taking. The model is trainable on consumer's-level GPU and runs in browser. It is a comprehensive practice for combining knowledge and experiences, offering a free and

### Overview: Jenetics is a comprehensive library for Genetic Algorithms, Evolutionary Algorithms, Grammatical Evolution, Genetic Programming, and Multi-objective Optimization written in modern Java. It simplifies the implementation of these algorithms by providing clear separation of concepts such as Gene, Chromosome, Genotype, Phenotype, and Population, and allows optimization without modifying the fitness function. ### Key Features: - **Genetic Algorithm (GA)** - **Evolutionary Algorithm** - **Grammatical Evolution (GE)** - **Genetic Programming (GP)** - **Multi-objective Optimization** ### Tech Stack: Java ### Why It Matters: Jenetics simplifies the implementation of genetic algorithms and related optimization techniques by providing a clear separation of concepts and a modular design. This makes it easier to implement and understand complex algorithms without the need to customize the fitness function. The library's use of an evolution stream allows for seamless integration with the Java Stream API, making it highly compatible with existing Java codebases. ### GitHub Stars: 910 ### Repository: Repository: [jenetics/jenetics](https://github.com/jenetics/jenetics) ### Primary Language: Java ### Build Jenetics: To build Jenetics, clone the repository and use the following commands: ```sh $ git clone https://github.com/jenetics/jenetics.git <builddir> $ cd <builddir> $ gradle build ``` ### Documentation: The library is fully documented with Javadoc and comes with a user manual in PDF format.

Overview: OpenFlowKit is a free, open-source diagram canvas that allows users to draw, write, or hand over diagrams to an AI model or local agent. It supports various diagramming languages such as Mermaid, D2, and Structurizr, and can map a GitHub repo to its corresponding code. Key features: - Supports multiple diagramming languages: Mermaid, D2, and Structurizr. - Live diagramming with an agent, where the agent reads the page, draws, and regenerates the diagram. - Text and canvas are the same diagram, allowing for easy editing and conversion between text and shapes. - Maps a GitHub repo to its corresponding code, integrating with the GitHub platform. Tech stack: TypeScript Why it matters: OpenFlowKit is a powerful tool for anyone looking to create and visualize diagrams, making it easier to understand complex systems and collaborate with others. Its integration with GitHub and live diagramming capabilities make it an invaluable asset for developers and designers alike.

### Overview: This repository is a collection of resources related to Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs). It provides implementations, tutorials, and examples for various applications of these neural network architectures. ### Key Features: - Implementations of Graph Convolutional Networks (GCNs) in TensorFlow, Keras, and PyTorch. - Implementations of Graph Attention Networks (GATs) in Keras and PyTorch. - Examples and tutorials on Graph Convolutional Neural Networks (GCNs). - Code for improved GCNs, such as Edge Attention-based Multi-relational Graph Convolutional Networks (EAGCN). - Knowledge Graphs, including implementations of Relational Graph Convolutional Networks (RGCNs) and Cross-lingual Knowledge Graph Alignment (KGAlign). - Relation Extraction, including implementations for improved relation extraction using pruned dependency trees and distant supervision. ### Tech Stack: Python, TensorFlow, Keras, PyTorch, and other relevant libraries. ### Why It Matters: These resources are essential for researchers and developers working on natural language processing, machine learning, and artificial intelligence. They provide practical tools and examples to enhance the performance of graph-based models, making it easier to implement and improve upon existing techniques in these fields.

### Overview: 3DMatch is a ConvNet-based local geometric feature descriptor designed for aligning 3D meshes and point clouds. This toolbox provides tools for geometric registration and keypoint matching, as well as training 3DMatch from existing RGB-D reconstructions. It is the reference implementation of our paper "3DMatch: Learning Local Geometric Descriptors from RGB-D Reconstructions". ### Key features: - Geometric Registration - Keypoint Matching - Training from RGB-D Reconstructions ### Tech stack: C++ ### Why it matters: 3DMatch addresses the challenges of matching local geometric features on noisy, low-resolution, and incomplete 3D scan data. By leveraging unsupervised feature learning from existing RGB-D reconstructions, it provides a robust and effective solution for tasks such as instance-level object model alignment and mesh surface correspondence. The results consistently outperform state-of-the-art methods, making it a valuable tool in the field of 3D data analysis and computer vision.

### Overview: This project implements the "Get To The Point: Summarization with Pointer-Generator Networks" model using PyTorch, focusing on improving the summarization process by incorporating pointer-generator networks. The model aims to enhance the quality of summaries by generating more relevant and concise text. ### Key Features: - **Pointer Generator Networks**: The project utilizes pointer-generator networks to generate summaries that are more focused and relevant. - **Coverage Loss**: The model incorporates a coverage loss to ensure that the generated summaries cover the most important parts of the input text. - **PyTorch Implementation**: The project is implemented using PyTorch, making it accessible and efficient for researchers and developers. ### Tech Stack: - **Programming Language**: Python - **Framework**: PyTorch ### Why It Matters: This project is significant because it provides a practical solution for improving the quality of text summarization, making it easier to generate summaries that are more focused and relevant. The use of PyTorch and pointer-generator networks ensures that the model is both effective and efficient, making it a valuable tool in the field of natural language processing. ### Repository: - **Repository**: atulkum/pointer_summarizer - **GitHub Stars**: 909 - **GitHub Description**: Pytorch implementation of "Get To The Point: Summarization with Pointer-Generator Networks" - **Primary Language**: Python - **GitHub Description**: Pytorch implementation of "Get To The Point: Summarization with Pointer-Generator Networks"