Projects

Overview: STEGO is an unsupervised semantic segmentation method that distills feature correspondences across images to achieve accurate object segmentation. Key features: - Unsupervised Semantic Segmentation - Distilling Feature Correspondences - Deep features connect objects across images - Results Tech stack: Jupyter Notebook Why it matters: This method offers a novel approach to unsupervised semantic segmentation by leveraging the deep features that connect objects across images, making it particularly useful for applications where labeled data is scarce. Repository: mhamilton723/STEGO Primary language: Jupyter Notebook GitHub stars: 791 GitHub description: Unsupervised Semantic Segmentation by Distilling Feature Correspondences README excerpt: [...] <!--ts--> * [Install](#install) * [Evaluation](#evaluation) * [Training](#training) * [Bringing your own data](#bringing-your-own-data) * [Understanding STEGO](#understanding-stego) * [Unsupervised Semantic Segmentation](#unsupervised-semantic-segmentation) * [Deep features connect objects across images](#deep-features-connect-objects-across-images) * [The STEGO architecture](#the-stego-architecture) * [Results](#results) * [Citation](#citation) * [Contact](#contact) <!--te-->

Overview: `neurodiffeq` is a Python library designed for solving differential equations using neural networks. It is based on PyTorch and has been adopted by multiple research groups worldwide, including at Harvard IACS. Key features: - Solves ordinary differential equations (ODEs) - Solves partial differential equations (PDEs) - Supports solution bundles - Can be used to solve reverse problems Tech stack: Python, PyTorch Why it matters: `neurodiffeq` offers a novel approach to solving differential equations by leveraging neural networks, providing a flexible and powerful tool for researchers and developers in various scientific and engineering fields.

### Overview: Phenaki Pytorch is an implementation of the Phenaki video generation technique, utilizing Mask GIT and CViViT for creating text-guided videos up to 2 minutes in length. This project leverages Pytorch for efficient processing and is designed to be replicable and scalable. ### Key Features: - **Phenaki Video Generation**: Produces videos from text prompts, up to 2 minutes in length. - **Mask GIT**: Uses a mask-based approach to generate videos from text descriptions. - **CViViT**: A convolutional video transformer for processing video frames. - **Mask Git**: A token-based model for generating coherent video sequences. ### Tech Stack: - Python - Pytorch ### Why It Matters: Phenaki Pytorch offers a practical and efficient way to generate coherent video content from textual descriptions, making it valuable for applications in media generation, education, and more. ### Repository: Repository: lucidrains/phenaki-pytorch GitHub Stars: 790 GitHub Description: Implementation of Phenaki Video, which uses Mask GIT to produce text guided videos of up to 2 minutes in length, in Pytorch.

Overview: This project is a comprehensive course designed to educate developers on the practical application of generative AI technologies, focusing on integrating them into cloud-based applications. Key features: - Cloud-based deployment - Use of OpenAI, Gemini, Streamlit, Containers, Serverless, Postgres, LangChain, Pinecone, and Next.js - Integration of AI into various enterprise systems - Scalable architecture - Integration with business domain experts - Training and validation of prompt engineering - Selection of appropriate LLMs based on performance and cost-effectiveness Tech stack: Python Why it matters: This course is crucial for developers as it equips them with the skills necessary to effectively integrate generative AI into their applications, making them essential for the future of cloud-based engineering.

### Overview: Claude-smart is a self-improvement plugin designed to enhance the skills of Claude Code, Codex, and OpenCode. It learns from your corrections and existing project-specific rules to create durable skills that Claude follows in future sessions. ### Key Features: - **Project-Specific Skills:** Captures rules from your projects. - **Shared Skills:** Rolls up proven patterns for reuse across different projects. - **Tech Stack:** Built on the open-source Reflexio engine. ### Why It Matters: Claude-smart improves the quality of skills generated by Claude Code, Codex, and OpenCode, making them more effective and reliable. It helps you achieve better results and maintain consistent performance across different projects. ### Repository: The repository for Claude-smart is located at <https://github.com/ReflexioAI/claude-smart>. It has been starred by 787 developers, indicating its popularity and utility in the community. ### Tech Stack: Claude-smart is built on the Reflexio engine, which is an open-source project. The plugin is compatible with Node.js 20+ and can be installed on your host CLI like Claude, Codex, or OpenCode. ### Quick Start: To install Claude-smart, use the following command: ```bash npx claude-smart install ``` You can choose between local Reflexio and managed Reflexio modes. The guide will help you set up the plugin, including where to store and process your data. ### How It Works: Claude-smart learns from your

### Overview: CodeGate is an experimental project designed to enhance the security, workspaces, and multiplexing capabilities of AI applications, coding assistants, and agentic frameworks. It provides a centralized environment for managing AI model and provider configurations, ensuring secure and efficient development processes. ### Key Features: - Centralize and manage AI coding assistants in one place. - Route prompts to different AI models based on workspaces or file types. - Centralize and manage provider credentials and API keys. - Protect your development process from accidental exposure of secrets and sensitive data. - Ensure recommendations follow secure coding practices. - Prevent the use of malicious or deprecated libraries. ### Tech Stack: - Python ### Why It Matters: CodeGate simplifies the management of AI model and provider configurations, providing a safer and more unified development environment. It ensures that AI-generated code adheres to best practices and maintains the integrity of your projects, making it an essential tool for developers working with AI applications and coding assistants.

### Overview: This repository is a curated collection of papers and resources related to the video generation task. It serves as a valuable resource for researchers, developers, and enthusiasts interested in the latest advancements in video synthesis and editing. ### Key Features: - **Video Generation Studies**: Includes papers and resources on various aspects of video generation, such as text-to-video, image-to-video, personalized video generation, video editing, and audio-to-video generation. - **Yearly Updates**: Papers and resources are regularly updated to reflect the latest research in each year. - **Code and Models**: Links to open-source code and models are provided for further exploration and experimentation. ### Tech Stack: TeX ### Why it Matters: This repository is essential for anyone looking to stay updated with the latest advancements in video generation research. It serves as a comprehensive resource for researchers, developers, and enthusiasts interested in this field. ### Repository: Repository: AlonzoLeeeooo/awesome-video-generation Stars: 787 Description: A collection of awesome video generation studies. ### Primary Language: TeX ### GitHub Stars: 787 ### GitHub Description: A collection of awesome video generation studies. ### Contents: - **To-Do Lists**: Updates for papers released in 2025, 2024, 2023, and 2022. - **Products**: Papers, surveys, text-to-video, image-to-video, personalized video generation, video editing, audio-to-video, and human image animation. - **Datasets

### Overview: This repository provides examples of Java applications built with Generative AI and Large Language Models (LLMs) using Spring AI and Spring Boot. It showcases various use cases such as chatbots, question answering, semantic search, structured data extraction, and text classification. ### Key Features: - **Chatbot**: Using LLMs via Ollama. - **Question Answering**: Answering questions with documents using LLMs via Ollama and PGVector. - **Semantic Search**: Semantic search using LLMs via Ollama. - **Structured Data Extraction**: Extracting structured data using LLMs via Ollama. - **Text Classification**: Classifying text using LLMs via Ollama. ### Tech Stack: Java 25, Podman/Docker. ### Why it Matters: This repository serves as a practical guide for developers looking to integrate Generative AI and LLMs into their Java applications, leveraging Spring AI and Spring Boot for a robust and efficient solution.

### Overview: Long Range Arena (LRA) is a benchmark designed to evaluate the efficiency and performance of transformer models. It provides a comprehensive set of tasks and datasets to assess the generalization power, computational efficiency, and memory footprint of various transformer variants. ### Key Features: - **Benchmark Tasks**: Includes a variety of tasks such as text classification, retrieval, and image classification. - **Variants of Transformers**: Implements different variants of transformers using JAX and Flax. - **Vanilla Transformers**: Provides the necessary code to run benchmarks on vanilla transformers. ### Tech Stack: Python ### Why It Matters: LRA serves as a standardized benchmark for evaluating the efficiency and performance of transformer models across different tasks and datasets, ensuring fair and comparable evaluations. ### Repository: - **Repository**: `google-research/long-range-arena` - **GitHub Stars**: 788 - **Description**: A benchmark for efficient transformers. - **Primary Language**: Python ### V2 Release: - **Model Releases**: The xformer models used in the experiments have been released. - **Future Updates**: Plans for additional models and baselines, with a 2nd update scheduled. ### Public External Entries: - **IGLOO**: Submissions by Vsevolod Sourkov. - **TLB**: Temporal Latent Bottleneck model. - **Model List**: Includes tasks like text classification, retrieval, and image classification. ### Citation: - **Citation**: If you find the work useful, please cite our paper at: ``` @inproceedings{

### Overview **june** is a local voice chatbot that leverages Ollama for language model capabilities, Hugging Face Transformers for speech recognition, and the Coqui TTS Toolkit for text-to-speech synthesis. It offers a privacy-focused solution for voice-assisted interactions on your local machine, ensuring no data is sent to external servers. ### Key Features - **Text Input/Output:** Provide text inputs to the assistant and receive text responses. - **Voice Input/Text Output:** Use your microphone to give voice inputs, and receive text responses from the assistant. - **Text Input/Audio Output:** Provide text inputs and receive both text and synthesised audio responses from the assistant. - **Voice Input/Audio Output (Default):** Use your microphone for voice inputs, and receive responses in both text and synthesised audio form. ### Tech Stack - Python 3.10 or greater with pip - PortAudio development package (for GNU/Linux) or Microsoft Visual C++ - Ollama, Hugging Face Transformers, Coqui TTS Toolkit ### Why It Matters **june** provides a flexible, privacy-focused solution for voice-assisted interactions on your local machine, ensuring no data is sent to external servers. It combines powerful language models, speech recognition, and text-to-speech capabilities to offer a seamless and engaging conversational experience. ### Repository Repository: **june** Primary Language: Python GitHub Stars: 788 GitHub Description: Local voice chatbot for engaging conversations, powered by Ollama, Hugging Face

### Overview: This project is a TensorFlow implementation of Character-Aware Neural Language Models, focusing on character-level language modeling. The model combines a convolutional neural network with a highway network and a recurrent neural network language model, aiming to enhance the performance of language modeling tasks. ### Key Features: - Word-level and Character-level Convolutional Neural Network - Highway Network - Recurrent Neural Network Language Model ### Tech Stack: Python ### Why it Matters: This implementation aims to improve upon the performance of previous models by incorporating character-level information into the language modeling process. The model's current performance issue is noted in the README, suggesting ongoing development and potential improvements.

Overview: Aphantasia is a project that utilizes CLIP and FFT/DWT/RGB parameterizers to generate detailed text-to-image and text-to-video tools. This innovative approach aims to overcome the limitations of traditional GANs, offering high-resolution and continuous processing capabilities. Key features: - Generating massive detailed textures similar to deepdream - FullHD/4K resolutions and above - Support for various CLIP models - Continuous mode for processing phrase lists - Pan/zoom motion with smooth interpolation - Direct RGB pixels optimization - 3D look based on Depth Anything 2 - Complex queries with text and/or image prompts - Option to subtract specific topics - Support for non-English languages with Google translation - Customizable options for composition, iteration count, and more Tech stack: Python Why it matters: Aphantasia addresses the limitations of traditional GANs by providing high-resolution and continuous processing capabilities, making it a valuable tool for creators and artists looking to generate detailed and visually appealing images and videos from text prompts. Repository: eps696/aphantasia GitHub stars: 788 GitHub description: CLIP + FFT/DWT/RGB = text to image/video

Overview: This project allows developers to utilize a microphone with OpenAI's Whisper model, providing a demo and various model sizes for different applications. Key features: - Microphone integration for speech-to-text conversion - Support for multiple model sizes and languages - Model weights released under the MIT License Tech stack: Python Why it matters: This project offers a cost-effective solution for developers to integrate speech recognition capabilities into their applications, potentially improving user experience and efficiency.

Overview: COMET is a neural framework designed for evaluating machine translation quality and error analysis. Key features: - Supports both reference-based and reference-free evaluation. - Provides quality scores and explanations for translation errors. - Includes a new method for extracting free-text explanations from XCOMET outputs. - Extends COMET to a document-level extension called DocCOMET, improving accuracy on discourse phenomena tasks. Tech stack: Python Why it matters: COMET offers a comprehensive solution for evaluating machine translation quality, providing detailed insights into translation errors and enabling developers to improve their models through error analysis. Repository: Unbabel/COMET GitHub stars: 785 GitHub description: A Neural Framework for MT Evaluation README excerpt: A Neural Framework for MT Evaluation

Overview: SwiftWhisper is an easy-to-use Swift package for transcribing audio files, powered by Whisper.cpp. It simplifies the process of adding speech recognition functionality to Swift applications. Key features: - Transcribes audio files into text segments - Supports both Swift Package Manager and Xcode integration - Provides delegate methods for handling transcription progress and errors - Includes pre-trained models for various languages Tech stack: Swift Why it matters: SwiftWhisper streamlines the process of integrating speech recognition into Swift applications, making it accessible and easy to use for developers. Repository: exPHAT/SwiftWhisper GitHub stars: 785

Overview: Hippo is a long-term memory system designed specifically for coding agents, improving their intelligence over time. It keeps data local and integrates with various tools like Claude Code, Codex, and Cursor, enhancing the overall performance and accuracy of AI agents. Key features: - **Memory Learning**: Hippo continuously learns from the outcomes of its interactions, improving its effectiveness over time. - **Local Storage**: Data is stored locally on the user's machine, ensuring privacy and security. - **Integration with Tools**: Hippo seamlessly integrates with tools like Claude Code, Codex, and Cursor, enhancing their functionality. Tech stack: TypeScript Why it matters: By preserving and refining memories, Hippo enhances the intelligence of coding agents, making them more efficient and accurate in their work. This system is critical for developers who need to maintain and improve their AI agents' performance without re-teaching them. Repository: kitfunso/hippo-memory Primary language: TypeScript GitHub stars: 784 GitHub description: Hippo is a long-term memory system for coding agents, improving their intelligence over time. It keeps data local and integrates with various tools, enhancing the overall performance and accuracy of AI agents.

### Overview: 本系列包括《数学基础》、《Python实践》、《NLP应用》、《工具与工程化》（含TensorFlow系列）、《行业应用与课程笔记》等多部分，旨在全面覆盖人工智能与深度学习的各个方面。通过实践操作和工具实践篇中的Jupyter Notebook，读者可以深入理解机器学习、深度学习、自然语言处理等技术的实际应用。 ### Key features: - 《数学基础》：涵盖数理统计、概率论等基础知识。 - 《Python实践》：提供Python编程实践，帮助读者掌握编程技能。 - 《NLP应用》：讲解自然语言处理的理论与实践。 - 《工具与工程化》：介绍TensorFlow、PyTorch等工具的使用方法。 - 《行业应用与课程笔记》：涵盖人工智能在医疗、金融、教育等领域的应用实例。 ### Tech stack: Jupyter Notebook ### Why it matters: 通过实践操作和工具实践篇中的Jupyter Notebook，读者可以深入理解机器学习、深度学习、自然语言处理等技术的实际应用。这些内容不仅有助于理论学习，还能通过实际操作加深对概念的理解，提高解决实际问题的能力。

Overview: RocketQA is a powerful toolkit for dense retrieval, providing state-of-the-art models and easy-to-use integration with JINA for development. Key features: - 🚀 State-of-the-art dense retrieval models - 🚀 First-Chinese model trained on DuReader - 🚀 Easy-to-use integration with JINA for quick development Tech stack: Python Why it matters: RocketQA simplifies the process of using cutting-edge dense retrieval models, making it accessible for developers, especially those interested in information retrieval and question answering.

Overview: RecSim is a configurable platform designed for authoring simulation environments for recommender systems (RSs), offering a way to naturally reflect user behavior and item structure in sequential interactive recommendation problems. It supports sequential interaction with users and allows for easy configuration of various aspects of user preferences, item familiarity, latent states, and choice models. Key features: - Configurable environments reflecting specific user behavior and item structure. - Support for sequential interaction with users. - Easy configuration of user preferences, item familiarity, latent states, and choice models. Tech stack: Python Why it matters: RecSim provides a valuable tool for researchers and practitioners in the field of recommendation systems, enabling them to push the boundaries of current reinforcement learning and recommendation system techniques through configurable simulation environments.

Overview: pycococreator is a Python library designed to simplify the creation of COCO datasets, providing functions for generating annotations in both uncompressed RLE and polygon formats. Key features: - Generates COCO-style annotations - Supports uncompressed RLE and polygons - Easy to integrate into existing projects Tech stack: Python Why it matters: pycococreator streamlines the process of creating COCO datasets, making it easier for developers to follow the COCO format and contribute to the community. Repository: waspinator/pycococreator Primary language: Python GitHub stars: 784