Projects

### Overview: Momy is a cutting-edge desk robot designed to enhance human-machine interaction with advanced features such as advanced vision, active listening, sound response, visual feedback, intelligent mobility, environmental monitoring, an optimized power system, and remote control capabilities. ### Key Features: - **Advanced Vision**: Integrated camera for perception and environment understanding. - **Active Listening**: Microphone for audio input and recognition. - **Sound Response**: Built-in speaker for vocal and sound communication. - **Visual Feedback**: Screen for visual communication. - **Intelligent Mobility**: Dynamic and interactive physical presence. - **Environmental Monitoring**: Thermometer for internal temperature and humidity monitoring. - **Optimized Power System**: Internal batteries and recharge management. - **Remote Control with ESP32S3**: Remote control and Internet browsing capabilities. ### Tech Stack: C ### Why it Matters: Momy is a versatile and intelligent robot designed to enhance daily interactions, offering advanced sensory and processing capabilities, making it an ideal companion for various activities. # GitHub Stars: 113 # Repository: momysnow/Momy-Desk-Robot # Primary Language: C # GitHub Description: Smart desktop robot. # [Documentation and Instructions](https://github.com/momysnow/Momy-Desk-Robot/wiki) # [Discord Server](https://discord.gg/V63W7n23pe) # Contributions: Welcome contributions from the community. Feel free to join the Discord server and propose your changes. # Copyright: All files in Momy-Desk-Robot by momysnow.

Overview: Morph is a Python-centric full-stack framework for building and deploying AI apps, offering fast setup, easy deployment, and no HTML/CSS knowledge required. Key features: - Fast to start - Deploy and operate - No HTML/CSS knowledge required - Chain Python and SQL for advanced AI workflows - Customizable components Tech stack: Python Why it matters: Morph simplifies the development of AI apps by providing a fast, easy-to-use framework that eliminates the need for HTML/CSS knowledge, making it accessible to developers without prior experience in these technologies.

Overview: This repository contains materials for a blog post on using Redis as a vector database with LlamaIndex and OpenAI for semantic search and chatbot-like interactions with PDF documents. Key features: - Jupyter Notebook: Demonstrates the use of Redis as a vector database and semantic search with LlamaIndex. - OpenAI API Integration: Shows how to use OpenAI for text and embedding model interactions. - Azure Redis Enterprise: Option to use Azure Redis Enterprise for advanced features. - Vector Similarity Search: Utilizes Redis Search & Query for efficient document retrieval. Tech stack: Jupyter Notebook Why it matters: This project provides a practical demonstration of how to integrate Redis with LlamaIndex and OpenAI for advanced search and chat functionalities, making it useful for developers looking to enhance their document management and search capabilities. Repository: redis-developer/LLM-Document-Chat GitHub stars: 113

Overview: ScribbleArchitect is a user-friendly application that leverages Stable Diffusion and ControlNet to transform simple scribbles into architectural designs, enhancing creativity and inspiration in the design process. Key features: - Real-time line drawing and segmentation from simple brush strokes or Bezier curves - Support for IP Adapter, ControlNet, and LCM for advanced style transfer and control - Customizable styles and predefined styles for various architectural styles - Upscale function for high-resolution output - Support for tablet drawing and drawing with color brushes Tech stack: Python Why it matters: ScribbleArchitect combines AI technology with traditional drawing techniques to streamline the architectural design process, making it easier and more efficient for architects and designers. Repository: s-du/ScribbleArchitect GitHub stars: 113

Overview: TalaDB is an open-source embedded vector and document database designed for building local-first AI applications. It combines a document database and a vector database into a single, embedded engine, enabling efficient querying and vector similarity search directly on the user's device. Key features: - Stores JSON-like documents and queries them using familiar filters. - Runs vector similarity search entirely on the user's device. - Built-in vector index for efficient vector similarity search. - Supports multiple platforms: browser, Node.js, React Native, Android (Kotlin), and iOS/macOS (Swift). Tech stack: Rust. Why it matters: TalaDB simplifies the development of local-first AI applications by providing a unified database engine that can be used across different platforms, reducing the need for a separate server and improving performance and security.

Overview: This project explores the development of a spoken keyword spotting (KWS) system using a hybrid approach combining a Convolutional Neural Network (CNN) and a Support Vector Machine (SVM). The goal is to create a low-memory footprint, efficient solution that can be used in always-on speech recognition applications. Key features: - Training and testing of a CNN model for feature extraction - Integration of SVM for classification - Implementation of a continuous speech streaming demo Tech stack: Python v3.8, TensorFlow v2.2 Why it matters: The project addresses the need for efficient and accurate speech recognition in always-on devices, reducing latency and improving privacy by leveraging local keyword detection followed by cloud-based speech recognition.

Overview: NineRec is a benchmark dataset suite designed for evaluating transferable recommendation systems across multiple domains and cross-domain scenarios. This project aims to provide a comprehensive evaluation platform for researchers in the field of recommendation systems. Key features: - Multimodal dataset for evaluating transferable recommendation models. - Supports end-to-end training of recommendation systems. - Includes a variety of tasks such as multi-modal, foundation model, transfer learning, and recommendation. - Provides a sample of 1000 images per dataset for easy access. Tech stack: Python Why it matters: NineRec offers a valuable resource for researchers to evaluate and compare different recommendation models across various domains and scenarios, promoting the development of more effective and efficient recommendation systems. Repository: westlake-repl/NineRec Primary language: Python GitHub stars: 113 GitHub description: Multimodal Dataset and Benchmark for Multi-domain and Cross-domain Recommendation System README excerpt: <p align="center" width="100%"> <img src='https://github.com/westlake-repl/NineRec/blob/main/assets/NineRec_logo_banner_v6-3.png' width="100%"> </p> # [TPAMI 2024] [NineRec: A Benchmark Dataset Suite for Evaluating Transferable Recommendation](https://arxiv.org/pdf/2309.07705.pdf) Feel free to contact us if you have questions regarding the database: yuanfajie@westlake.edu.cn <a href="https://arxiv.org/pdf/2

Overview: This project implements a distributed AI system that runs a 56M-parameter language model across three ESP32-S3 microcontrollers using ESP-NOW communication, enabling fully offline inference. Key features: - Distributed inference across three ESP32-S3 boards - Split-PLE architecture with a 4-bit quantized PLE table and a KV cache for efficient token processing - Web-based interaction for user-friendly interface Tech stack: C Why it matters: This project demonstrates the feasibility of distributed AI inference on low-power microcontrollers, showcasing the potential for real-time AI processing in embedded devices. Repository: wladimiravila/esp32s3-distributed-ai GitHub stars: 113

Overview: Target-Driven Distillation (TDD) is a state-of-the-art consistency distillation model designed to accelerate inference processes of diffusion models. By employing *target timestep selection* and *decoupled guidance*, TDD can generate highly detailed images with minimal steps, significantly outperforming previous methods like LCM, PCM, and TCD. Key features: - TDD is a consistency distillation model that accelerates inference processes of diffusion models. - It uses *target timestep selection* to control the inference steps. - TDD employs *decoupled guidance* to improve the clarity and detail of generated images. Tech stack: Python Why it matters: TDD offers a significant improvement in the efficiency and quality of diffusion model inference, making it a valuable tool for researchers and practitioners working with diffusion-based models.

Overview: This repository contains the codes for our paper "Conditional Antibody Design as 3D Equivariant Graph Translation", focusing on the development of a multi-channel, equivariant attention network (MEAN) for conditional antibody design. The project aims to enhance the design of antibodies with improved accuracy and efficiency. Key features: - Multi-channel equivariant attention network (MEAN) - Conditional antibody design as 3D equivariant graph translation Tech stack: Python Why it matters: The MEAN project is significant for its innovative approach to antibody design, offering enhanced accuracy and efficiency compared to traditional methods. This technology could significantly impact the field of biotechnology and drug development. Repository: THUNLP-MT/MEAN Primary language: Python GitHub stars: 112 GitHub description: This repo contains the codes for our paper Conditional Antibody Design as 3D Equivariant Graph Translation.

Overview: TI MINDMAP GPT is an AI-powered tool designed to help produce Threat Intelligence Mindmaps, streamlining the process of extracting key insights from various sources. Key features: - Multiple input sources: Analyze content from URLs, uploaded PDF files, or direct text input. - Leverages OpenAI, Azure OpenAI, and MistralAI for text reduction and theme encapsulation. - Supports 24/7 automated processing, 50+ OSINT sources, and a fully hosted infrastructure. - Provides STIX 2.1 export and cross-report search engine capabilities. - Supports weekly briefings and AI-powered multi-agent analysis. - Offers community-driven enrichment for article submissions. - Supports real-time analytics dashboard and MCP Server, Knowledge Graph, and PDF Reports. - Supports a community-driven knowledge base (KB) for chat interactions. - Supports a knowledge graph for better understanding of the data. - Supports a multi-agent analysis for comprehensive threat intelligence. - Supports a community-driven knowledge base (KB) for chat interactions. - Supports a knowledge graph for better understanding of the data. - Supports a multi-agent analysis for comprehensive threat intelligence. - Supports a community-driven knowledge base (KB) for chat interactions. - Supports a knowledge graph for better understanding of the data. - Supports a multi-agent analysis for comprehensive threat intelligence. - Supports a community-driven knowledge base (KB) for chat interactions. - Supports a knowledge graph for better understanding of the data. - Supports a multi-agent analysis for comprehensive threat intelligence. - Supports a community-driven knowledge base (KB) for chat interactions. -

### Overview: DiffusionPen is an official PyTorch implementation of "DiffusionPen: Towards Controlling the Style of Handwritten Text Generation" presented at ECCV 2024. This model is designed to generate stylized handwritten text using a few reference samples, effectively capturing both seen and unseen handwriting styles. ### Key Features: - **Few-Shot Generation**: DiffusionPen can generate new text that closely resembles a writer's unique handwriting style using just a few reference samples. - **Style Extraction Module**: A style extraction module combines metric learning and classification to represent various writing styles effectively. - **Evaluation**: The model was evaluated on IAM and GNHK handwriting datasets, demonstrating its ability to generate diverse and realistic handwritten text. ### Tech Stack: Python ### Why it Matters: DiffusionPen offers a significant improvement in the style control of handwriting generation, making it a valuable tool for enhancing handwriting text recognition (HTR) systems. Its ability to generate realistic and diverse text styles with fewer samples makes it a powerful addition to the field of text generation and recognition.

### Overview: This TypeScript API, built with Hono, OpenAI, and deployed as a Cloudflare Worker, aims to enhance the performance of applications by integrating generative AI capabilities seamlessly. It allows developers to supercharge their applications with AI-powered shortcuts, providing a simple and easy-to-use interface. ### Key Features: - **Integration with Cloudflare Workers**: Seamlessly integrates generative AI capabilities into applications. - **Deployment and Configuration**: Deployed as a Cloudflare Worker, requires minimal configuration. - **API Endpoint**: `https://your-cloudflare-worker-url.com/api/v1/completion` for accessing the API. - **Providers & Models**: Supports providers like `google-ai-studio` and models like `gemini-2.5-flash`, `gemini-2.5-flash-lite`, `gemini-2.0-flash`, and `gemini-2.0-flash-lite`. ### Tech Stack: - TypeScript - Hono - OpenAI - Cloudflare Workers ### Why It Matters: This API simplifies the integration of generative AI into applications, making it accessible and easy to use for developers. It leverages the power of generative AI to create more intelligent and powerful shortcuts, enhancing the user experience without requiring extensive coding knowledge. ### Repository: - **GitHub Repository**: [louisbrulenaudet/genai-api](https://github.com/louisbrulenaudet/genai-api) - **GitHub Stars**: 111 - **GitHub Description**: Supercharge your applications using Cloudflare Workers and Gemini within minutes

### Overview: This repository is a curated list of strategies, tools, papers, and resources designed to reduce the cost of large language models (LLMs) and enhance their efficiency in production environments. ### Key features: - **Prompt Caching**: 90-97.5% off repeated prefix tokens, with verified savings and where each number comes from. - **Batch APIs**: 50% off input and output, with stacks that reduce input cost by up to 95-98.75%. - **Tool-Output Compression**: 20-35% of coding-agent input, with measured sessions showing improvements. - **Tool Search + Programmatic Tool Calling**: Over 85% off a tool-schema prefix, with reduced input tokens on a benchmark. - **Model Routing**: 2x+ cost cut at matched quality, with routes based on task complexity. ### Tech stack: Shell ### Why it matters: Reducing LLM token costs by 80-99% without sacrificing quality is crucial for cost-effective and efficient use of large language models in production environments.

### Overview: MarryWise-LLM: AI-Powered Suitor Analysis ### Key features: - **AI Matchmaking Analyst**: Analyzes behavior and speech patterns to uncover hidden motives and true personality traits. - **AI Fake Matchmaker Detector**: Identifies deceitful behavior to ensure the authenticity of potential partners. ### Tech stack: Python ### Why it matters: MarryWise-LLM uses advanced AI technology to help users make informed decisions about potential partners, ensuring a more authentic and reliable match.

Overview: Ssebowa is an open-source and free-to-use generative AI platform that offers access to powerful AI models for everyone, serving humanity. It provides text generation, image creation, and image understanding functionalities. Key features: - Free-to-use AI platform - Access to powerful AI models - Text generation - Image creation from text descriptions - Image understanding Tech stack: TypeScript Why it matters: Ssebowa aims to democratize AI by providing access to cutting-edge technology without financial barriers, making it accessible to everyone, including children in need.

Overview: Conversify is a real-time, low-latency AI assistant built on LiveKit, featuring local LLM integration and conversational memory. Key features: - Low latency response time under 600 ms. - Real-time voice and vision capabilities. - Compatible with OpenAI-style APIs. - Local LLM integration. - Basic vision processing. - Persistent conversational memory. Tech stack: Python. Why it matters: Demonstrates highly responsive conversational AI workflows using locally hosted models, making it suitable for applications requiring real-time interaction and low latency. Repository: taresh18/conversify GitHub stars: 112 GitHub description: 🗣️ Real‑time, low‑latency voice, vision, and conversational‑memory AI assistant built on LiveKit and local LLMs.

### Overview: UniMedVL is a unified medical vision-language model that excels in understanding and generating medical information across various modalities. It achieves state-of-the-art performance on multiple evaluation datasets, including 75.4% accuracy on SLAKE V. ### Key Features: - **Unified Observation-Knowledge-Analysis Architecture**: Enables seamless multimodal understanding and generation within a single architecture. - **Versatile Medical Multimodal Capabilities**: Supports visual question answering, medical report generation, text-to-medical-image synthesis, cross-modal translation, and virtual staining across 9 imaging modalities. - **Large-Scale Medical Dataset**: UniMed-5M, a comprehensive dataset with 5.6M+ high-quality samples, is used for training and evaluation. ### Tech Stack: Python ### Why it Matters: UniMedVL is a significant advancement in medical AI, providing a unified approach to understanding and generating medical information across various modalities. Its superior performance and comprehensive dataset make it a valuable tool for researchers and practitioners in the field of medical AI. Repository: uni-medical/UniMedVL GitHub Stars: 112

### Overview: This repository is a comprehensive list of research papers focusing on Large-Language-Diffusion-Models (dLLMs). It covers various methodologies, reasoning, and optimization techniques, as well as system efficiency and applications. The papers are categorized into different sections, including surveys, core methodologies, reasoning and policy optimization, token ordering and generation strategies, system efficiency and acceleration, multi-modal and physical AI, agentic and tool-use dLLMs, theory, guidance, and applications, and seminal diffusion papers. ### Key Features: - Surveys and useful resources - Core methodologies: Discrete & Masked Diffusion, Continuous & Latent Space Diffusion, AR-to-Diffusion Adaptation, Hybrid AR-Diffusion - Reasoning & Policy Optimization: Reasoning & Planning, Alignment & Reinforcement Learning - Token Ordering & Generation Strategies - System Efficiency & Acceleration: Caching & Memory Strategy, Decoding & Sampling, Distillation, Quantization & Sparsity, Inference Frameworks & Systems, Speculative Decoding & Draft Verification - Multi-modal & Physical AI: Multi-modal dLLMs, Vision-Language-Action (VLA), Autonomous Driving / World Models - Agentic & Tool-Use dLLMs - Theory, Guidance, and Applications: Theory & Analysis, Guidance & Downstream Applications - Seminal Diffusion Papers ### Tech Stack: Unknown ### Why It Matters: This repository serves as a valuable resource for researchers, developers, and enthusiasts interested in the latest advancements in Large-Language-Diffusion-Models. It

Overview: This project introduces a large Chinese language-and-vision assistant designed specifically for biomedical research, leveraging the capabilities of LLaVA-Med, a model that combines language understanding with medical image analysis. Key features: - Integrates language and vision capabilities for medical applications - Utilizes LLaVA-Med model for advanced biomedical research tasks - Provides detailed analysis tools for medical imaging datasets Tech stack: Python Why it matters: This project offers a powerful tool for biomedical researchers, enabling them to analyze medical images and documents with high accuracy, thereby enhancing the efficiency and effectiveness of their research. Repository: BUAADreamer/Chinese-LLaVA-Med GitHub stars: 111 GitHub description: 中文医学多模态大模型 Large Chinese Language-and-Vision Assistant for BioMedicine