Projects

Overview: RushDB is a powerful, schema-free graph and vector database designed specifically for AI agents and applications. It allows you to push any JSON data, get typed and searchable records, and automatically infer relationships and semantic search capabilities. Key features: - Push any JSON data - Get typed and searchable records - Automatically infer relationships and semantic search capabilities - No schema required, no migrations needed - Built on Neo4j Tech stack: TypeScript Why it matters: RushDB simplifies the process of integrating AI agents and applications by providing a single API for handling JSON data, graph traversal, and semantic search. It eliminates the need for separate databases and pipelines, making it easier to manage and scale AI solutions. Repository: rush-db/rushdb GitHub stars: 325

### Overview: This repository is the official codebase of "Let's Think Outside the Box: Exploring Leap-of-Thought in Large Language Models with Creative Humor Generation," accepted at CVPR 2024. The project aims to explore the "Leap-of-Thought" (LoT) ability in large language models through a humor generation game called Oogiri, challenging LLMs to think outside the box. ### Key features: - **Creative Humor Generation**: The project focuses on generating humorous responses to multimodal information. - **Oogiri Game**: Utilizes the Oogiri game to explore the LLM's LoT ability. - **Zero-Shot Inference**: Provides a simple Chinese example in `inference.py` for zero-shot inference. ### Tech stack: - Python ### Why it matters: This project is significant as it is the first to deeply explore the LoT ability in multimodal LLMs, challenging LLMs to think outside the box and generate unexpected and humorous responses. ### GitHub stars: 326

Overview: DiT-3D is an innovative model that leverages Plain Diffusion Transformers to generate 3D shapes directly from voxelized point clouds, offering a novel approach to 3D shape generation. Key features: - Uses Plain Diffusion Transformers for 3D shape generation - Supports multiple configuration settings including voxel sizes, patch dimensions, and model complexity - Provides pre-trained models for easy experimentation and comparison Tech stack: Python Why it matters: DiT-3D offers a unique method for 3D shape generation, potentially revolutionizing the field by enabling direct 3D shape generation from point clouds, which could simplify the process and potentially improve efficiency. Repository: DiT-3D/DiT-3D GitHub stars: 325 GitHub description: Official codebase of "DiT-3D: Exploring Plain Diffusion Transformers for 3D Shape Generation" README excerpt: 🔥🔥🔥DiT-3D is a novel Diffusion Transformer for 3D shape generation, which can directly operate the denoising process on voxelized point clouds using plain Transformers. [**DiT-3D: Exploring Plain Diffusion Transformers for 3D Shape Generation**](https://arxiv.org/abs/2307.01831) | [**Paper**](https://arxiv.org/abs/2307.01831) | [**Hugging Face**](https://huggingface.co/papers/2307.0183

### Overview: MentaLLaMA is an open-source instruction-following large language model designed for interpretable mental health analysis. This repository introduces the first open-source tool for analyzing mental health on social media using large language models. ### Key Features: - **Interpretability**: Provides clear explanations and insights into mental health analysis. - **Mental Health Analysis**: Analyzes mental health data from various social media platforms. - **Large Language Models**: Utilizes large language models to interpret and analyze mental health-related text. ### Tech Stack: Python ### Why it Matters: This project offers a practical solution for understanding mental health through social media, providing valuable insights without the need for clinical diagnosis.

### Overview: RadioDiff is an effective generative diffusion model designed for sampling-free dynamic radio map construction. This project addresses the challenge of obtaining radio maps without expensive sampling, making it highly relevant for 6G network applications. ### Key Features: - **RadioDiff:** The foundational diffusion model for radio map construction. - **RadioDiff-k²:** PINN-enhanced diffusion guided by the Helmholtz equation. - **iRadioDiff:** Indoor radio map construction with physical information integration. - **RadioDiff-Turbo:** Efficiency-enhanced RadioDiff for accelerated inference. - **RadioDiff-Flux:** Adaptive reconstruction under dynamic environments and base station location changes. - **RadioDiff-3D:** 3D radio map construction with the UrbanRadio3D dataset. - **RadioDiff-FS:** Few-shot learning for radio map construction with limited measurements. - **RadioDiff-Inverse:** Sparse measurement-based radio map recovery for ISAC applications. - **RadioDiff-Loc:** Sparse measurement-based NLoS localization using diffusion models. ### Tech Stack: Python ### Why It Matters: This project provides a practical solution for radio map construction, which is crucial for reducing communication costs in 6G networks. The use of generative diffusion models makes it possible to construct radio maps without the need for expensive sampling, making it highly valuable for real-world applications. Repository: UNIC-Lab/RadioDiff GitHub Stars: 324 GitHub Description: This is the code for the paper "RadioDiff: An Effective Generative Diffusion Model for Sampling-Free Dynamic Radio

### Overview: SelfCodeAlign is the first fully open and transparent pipeline that enhances a code language model without relying on human annotations or distilled data from large, proprietary models. This approach led to the creation of StarCoder2-Instruct, a fully transparent, permissively licensed, self-aligned code model that achieves state-of-the-art performance in coding tasks. ### Key Features: - **Self-Alignment Pipeline:** Enhances a code language model without human annotations or proprietary data. - **StarCoder2-Instruct:** The first entirely self-aligned code LLM trained with a fully permissive and transparent pipeline. - **Model:** [bigcode/starcoder2-15b-instruct-v0.1](https://huggingface.co/bigcode/starcoder2-instruct-15b-v0.1) - **Code:** [bigcode-project/starcoder2-self-align](https://github.com/bigcode-project/starcoder2-self-align) - **Dataset:** [bigcode/self-oss-instruct-sc2-exec-filter-50k](https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-exec-filter-50k/) ### Tech Stack: - Python ### Why It Matters: SelfCodeAlign and StarCoder2-Instruct represent a significant advancement in the field of code generation by providing a fully transparent and permissive approach to model training and fine-tuning, leading to improved performance and accessibility.

### Overview LangConnect Client is a modern GUI interface designed for managing vector databases, leveraging PostgreSQL with the pgvector extension. It includes features for CRUD operations, document management, advanced search capabilities, authentication, and integration with AI assistants through the Model Context Protocol (MCP). ### Key Features - **Collection Management**: CRUD operations with custom metadata support. - **Document Management**: Multi-format support, automatic text extraction, and batch upload. - **Advanced Search**: Semantic search with OpenAI embeddings, keyword search with PostgreSQL, and hybrid search. - **Authentication**: Supabase JWT authentication with automatic token refresh and role-based access control. - **MCP Integration**: Integration with 9+ AI assistants (Claude, Cursor) and support for stdio and SSE transport. - **Modern UI**: Built with Next.js and Tailwind CSS, featuring dark/light themes and multi-language support (English/Korean). ### Tech Stack - **TypeScript**: The primary language used for development. - **Next.js**: The frontend framework powering the application. - **TypeScript**: The backend framework for handling API requests and responses. - **PostgreSQL**: The database powering the vector database management. - **pgvector**: The extension for PostgreSQL to support vector database management. - **Supabase**: The authentication and database management service. - **FastAPI**: The backend API for handling requests and responses. - **Python**: The scripting language used for backend tasks. - **FastAPI**: The frontend framework for the GUI interface. - **PostgreSQL**: The database powering the vector database management. -

**Overview:** Afterglow is a locally hosted AI companion that leverages real chat records to build and retrieve relationship memories, personas, and expression styles. It uses RAG (Reinforcement Learning from Human Dialogue) to maintain familiar conversational patterns in new interactions. **Key features:** - **Chat Record Integration:** Cleans, splits, and vectorizes real chat logs. - **Persona and State Management:** Combines persona, circadian patterns, life state, and relationship memories. - **Retrieval Paths:** Uses five vector paths across text history, image descriptions, and live memory. - **Optional Rewrite and Rerank:** Enhances retrieval accuracy with optional rewriting and reranking. **Tech Stack:** Python **Why it Matters:** Afterglow provides a way to maintain the warmth of past conversations in new interactions, making it a valuable tool for those seeking to preserve and continue the emotional connection from past conversations. **GitHub Stars:** 324

### Overview: LoRA Dataset Studio V2 is a self-hosted, one-tab workbench designed for the entire training lifecycle of image datasets. It allows users to build, curate, caption, clean, and train datasets using a single interface. The platform supports character, concept, and style workflows, image and ZIP/folder imports, subject-aware shot catalogs, local Klein and Krea 2 Edit generation, and more. ### Key Features: - Character, Concept, and Style workflows. - Image and ZIP/folder import. - Subject-aware shot catalogs. - Local Klein and Krea 2 Edit generation. - Reference editing and exact retries. - Captioning with local vision models or JoyCaption. - Image and `.txt` image/`.txt` round trips. - Guided ai-toolkit recipes for Z-Image, SDXL, Krea 2, FLUX.1, FLUX.2 Klein, Anima, and Qwen-Image 2.1. - Queues and advanced settings for checkpoint continuation. - Experimental slider LoRAs for LoRA training. - Generate with all seven image training families. - Fixed-seed checkpoint/strength comparisons, prompt batches, runs, logs, lineage, votes, rankings, and a generated-image Gallery. - Backup/restore without API keys. - ComfyUI deployment. - Configurable storage and Trash. ### Tech Stack: - Python ### Why It Matters: LoRA Dataset Studio V2 offers a comprehensive, user-friendly solution for building, curating, and training image datasets. It streamlines the entire

Overview: Vector AI is a platform designed to simplify the development of vector-based applications, enabling quick and easy creation, storage, manipulation, search, and analysis of vectors alongside JSON documents. Key features: - Multimedia Data Vectorisation: Images, Audio, and more can be turned into vectors. - Document Oriented Store: Vectors and documents can be stored together without needing metadata lookup. - Vector Similarity Search: Enables powerful vector-based search capabilities. - Hybrid Search: Combines vector search with traditional search features for enhanced functionality. - Multi-Model Weighted Search: Allows for flexible searches with different model weights. - Vector Operations: Supports flexible operations on vectors. - Aggregation: Offers traditional aggregation features. - Clustering: Helps interpret vectors and data by bucketing and providing statistics. - Vector Analytics: Provides practical vector analytics for better understanding of vectors. Tech stack: Python Why it matters: Vector AI simplifies the process of building vector-based applications, making it easier and faster to create, store, and analyze data using vectors alongside JSON documents.

Overview: MRLib is an open-source library designed for advanced multimodal recommendation models. Key features: - Automatic modality discovery - Intelligent graph caching - Real-time visualization - Continuous updates - Extensive dataset support Tech stack: Python Why it matters: MRLib provides a comprehensive benchmark and code base for multimodal recommendation, making it a valuable resource for researchers and developers in the field.

Overview: DeepSeek CLI is a command-line interface (CLI) tool that utilizes the DeepSeek Coder models to enhance development workflows and coding experience. It offers features such as intelligent code completion, analysis of large codebases, modernizing legacy code, debugging, and project scaffolding across multiple programming languages. Key features: - Code Completion & Generation - Repository-Level Understanding - Code Refactoring & Migration - Debugging & Code Review - Project Scaffolding - Multi-Language Support Tech stack: TypeScript Why it matters: DeepSeek CLI streamlines development by providing powerful tools to accelerate workflows, enhance code quality, and modernize projects efficiently. Its ability to work across multiple languages makes it highly versatile and beneficial for developers across various coding environments.

### Overview: 这个仓库是一个综合资料合集，涵盖了人工智能算法方面的求职面试、机器学习、深度学习和强化学习等方面的内容。它提供了大量资料和代码，帮助开发者准备面试和学习相关技术。 ### Key features: - 算法面试题汇总 - 大厂内推实习/春招/社招信息 - 机器学习、深度学习和强化学习的资料和代码 - Github、牛客网、知乎、博客、微信公众号和其它资源的面试题和答案 ### Tech stack: - Github ### Why it matters: 这个仓库为开发者提供了丰富的资料和代码资源，帮助他们准备面试和学习人工智能算法相关技术。

Overview: This repository contains various models and code for paraphrase identification tasks, specifically focusing on the Quora Question Pairs dataset. The project implements different models such as a basic Siamese LSTM baseline, a Siamese LSTM model with an added matching layer, and the Bilateral Multi-Perspective Matching model. The code is documented to provide a didactic example for TensorFlow users. Key features: - A basic Siamese LSTM baseline - A Siamese LSTM model with an added matching layer - The Bilateral Multi-Perspective Matching model Tech stack: Python Why it matters: This project serves as a resource for those interested in learning and implementing natural language processing models, specifically for paraphrase identification tasks. The provided code and documentation make it an excellent example for beginners and intermediate users of TensorFlow. GitHub stars: 324 Repository: nelson-liu/paraphrase-id-tensorflow

### Overview: This repository is a curated collection of datasets and resources dedicated to legal text processing. It includes a diverse range of datasets and tools related to the legal domain, grouped by region and jurisdiction. The resources are publicly accessible, with most being open-source and available for free. ### Key Features: - **WorldLII**: A federated gateway to free legal information worldwide, covering legislation, case law, treaties, and more. - **vLex**: A global legal research platform with case law, legislation, journals, and analytics. - **Westlaw**: Thomson Reuters' flagship global legal research service, offering case law, statutes, and secondary sources. - **LexisNexis / Lexis+**: Global legal content with case law, statutes, Practical Guidance, and news. - **HeinOnline**: Extensive law journal and historical legislative archives, treaties, and session laws. - **WIPO Lex**: A global database of IP laws and treaties, including WIPO Lex-Judgments. - **SSRN (Legal Scholarship Network)**: An open repository of legal scholarship and preprints. - **OpenAlex**: An open repository of scholarly metadata and abstracts, useful for legal literature mining. - **Pile-of-Law**: A 256GB open corpus of English legal/administrative text for legal NLP. - **MultiLegalPile**: A multilingual/multijurisdiction legal text corpus for large language models. - **LexGLUE**: A benchmark suite for legal language understanding across multiple domains. - **LEDGAR**: A dataset

### Overview: This repository serves as a curated collection of research papers and datasets focused on Human-Interaction-Motion-Generation applications. It aims to provide a comprehensive reference for researchers and practitioners in the field, emphasizing the importance of capturing the complexity of human dynamics in interactive settings. ### Key Features: - **Papers**: Over 30 papers on Human-Interaction-Motion-Generation, including papers on Human-Human Interaction, Human-Object Interaction, and Human-Scene Interaction. - **Datasets**: Various datasets related to Human-Interaction-Motion-Generation, covering different types of interactions such as human-human, human-object, and human-scene interactions. - **Code**: Papers and datasets are organized into distinct categories, making it easier to find relevant content. ### Tech Stack: The repository uses Markdown and GitHub to structure and organize the content. ### Why it Matters: This project is crucial for researchers and practitioners in the field of Human-Interaction-Motion-Generation, as it provides a structured reference that can guide advancements and applications in robotics, virtual reality, and animation.

### Overview: GAIA is a proactive personal AI assistant that automates your daily tasks and integrates with popular tools like Gmail, Calendar, Slack, and more. It handles everything from triaging emails to prepping for meetings, all without any setup required. ### Key features: - **Proactive**: Connects to your tools and starts automations immediately. - **Workflows**: Automations run on schedule or triggered by events. - **Memory**: Learns from conversations and preferences. - **Voice**: Real-time calls and a "Hey GAIA" wake word. - **Multi-platform**: Available on iMessage, WhatsApp, Telegram, Slack, Discord, web, desktop, and mobile. - **Integrations**: Connects to services via an MCP server. - **Code execution**: A sandboxed workspace for PDFs, decks, and spreadsheets. - **Deep research**: Multi-source web research with structured output. - **Browser use**: Drives a real browser to complete tasks on any website. - **Skills**: Built on the open Agent Skills spec and can be installed from GitHub. - **One workspace**: Inboxes, calendars, todos, and notifications in a single app. - **Self-hostable**: Open source, keys, models, no caps. ### Why it matters: GAIA streamlines your daily tasks, saving you time and effort. It handles everything from email triaging to meeting prep, all without any setup required. By connecting to your favorite tools, GAIA integrates seamlessly and runs the moment you connect, making your day easier and more efficient.

Overview: This project is a code repository for the paper "Binding Language Models in Symbolic Languages" presented at ICLR 2023. It provides a Binder environment for language models, showcasing state-of-the-art performance with minimal program annotations. Key features: - Binder environment for language models - Supports OpenAI chat series models like Codex - Demonstrates high performance with only dozens of program annotations Tech stack: Python Why it matters: This project showcases innovative techniques in binding language models to symbolic languages, potentially enabling more efficient and effective AI applications in symbolic reasoning and symbolic AI. Repository: xlang-ai/Binder GitHub stars: 324

### Overview: OpenClaw Assistant is a native Android voice client for OpenClaw and Hermes Agent, designed to integrate seamlessly with existing OpenClaw features. It supports wake word activation, system-level controls, and Wear OS support, offering a robust and user-friendly experience. ### Key Features: - **Wake Word Activation**: Automatically recognizes and responds to wake words. - **System-Level Controls**: Provides full system-level access and controls. - **Hermes Agent Integration**: Seamlessly integrates with the Hermes Agent backend, allowing for real-time chat and continuous conversation. - **Mobile Bridge**: Exposes a local HTTP service for Hermes to reach curated Android capabilities. - **Multi-Backend Support**: Supports OpenClaw Gateway, OpenClaw HTTP, and Hermes API Server as backend options. - **Primary Backend Configuration**: Existing OpenClaw Assistant installs upgrade in place, migrating existing settings to the new multi-backend repository. - **Hermes Setup in 30 Seconds**: Easy setup with just a few steps, including authentication and API key configuration. ### Tech Stack: Kotlin ### Why It Matters: OpenClaw Assistant offers a powerful and integrated solution for Android users, providing a native experience with wake word activation, system-level controls, and seamless integration with the OpenClaw and Hermes Agent platforms. This makes it ideal for anyone looking to enhance their Android device with a more advanced voice assistant. ### Repository: - **Repository**: yuga-hashimoto/OpenClaw-Assistant - **GitHub Stars**: 324 - **Primary Language**: Kotlin - **GitHub Description

Overview: ChatPLUG is a Chinese open-domain dialogue system aimed at building and sharing a Chinese personalized large language model. Key features: - Knowledge Augmentation: Integrates external knowledge during inference, allowing for up-to-date information or domain knowledge. - Personalization: Customizable conversation style and characters through bot profiles or role-play instructions. - Multi-Task Abilities: Displays proficiency in open-domain dialogue and impressive abilities on a wide range of NLP tasks. Tech stack: Python Why it matters: ChatPLUG is designed to enhance the integration of external knowledge during inference, making it versatile for various applications. Its personalization features allow for easy customization, making it suitable for both research and practical use.