Projects

### Overview: QuantaLogic: Unleash AI for Coding, Automation, and Conversations QuantaLogic is your go-to framework for building smart agents that code, automate workflows, and chat like pros. Powered by large language models (LLMs) and a versatile toolset, it offers three killer modes: **ReAct** for tackling tough tasks, **Flow** for streamlined processes, and **Chat** for natural, tool-savvy conversations. Whether you’re a coder, a business innovator, or an AI enthusiast, QuantaLogic delivers fast, flexible, and fun solutions. Let’s blast off! ### Key Features: - **ReAct Mode**: Solve problems with LLM reasoning + tool actions. - **Flow Mode**: Craft structured workflows with nodes and transitions. - **Chat Mode**: Converse naturally with tool-calling smarts. - **LLM Integration**: Supports OpenAI, Anthropic, DeepSeek via LiteLLM. - **Toolset**: Code execution, web search, file management, and custom tools. - **Smart Memory**: Keeps context lean for long tasks or chats. - **Real-Time Insights**: Track progress with events and logs. - **Safe Execution**: Docker-based tool isolation. ### Tech Stack: Python ### Why It Matters: QuantaLogic is your creative sidekick, saving time and unlocking possibilities. From coding scripts to automating business tasks or chatting about the universe, QuantaLogic is your practical magic, making AI work for you. - **Versatile Power**: Code, automate, or converse—handle any task.

Overview: This repository contains the code and documentation for a semantic search engine and RAG chatbot project that utilizes Upstash Vector and Upstash RAG Chat SDK to index and search over 144 million vectors from Wikipedia articles in 11 languages. Key features: - Indexed over 144 million vectors from Wikipedia articles in 11 languages - Used BGE-M3 embedding model for multilingual support - Implemented semantic search with cross-lingual capabilities - Created a RAG chatbot using Upstash RAG Chat SDK Tech stack: TypeScript Why it matters: This project showcases the capabilities of Upstash Vector and RAG Chat SDK by demonstrating the indexing and semantic search of Wikipedia articles across multiple languages, highlighting the project's potential applications in semantic search and language processing.

### Overview: This repository contains a collection of recent video understanding datasets, focusing on various aspects of video classification, temporal action detection, spatio-temporally localized atomic visual actions, hand gestures, video captioning, and video question answering. The datasets are categorized by their specific tasks and organizers, providing valuable resources for researchers and developers in the field of computer vision and artificial intelligence. ### Key Features: - **Video Classification**: Includes datasets like UCF101, HMDB51, ActivityNet v1.3, and more, focusing on human actions and activities. - **Temporal Action Detection**: Features datasets such as THUMOS2014, ActivityNet(v1.3), and Broad Video Highlights, aimed at detecting temporal actions. - **Spatio-Temporally Localized Atomic Visual Actions**: Contains datasets like AVA, focusing on spatio-temporally localized visual actions. - **Hand Gestures in Videos**: Features datasets like Jester, which focus on hand gestures in videos. - **Video Captioning**: Includes datasets like MPII-MD, MSR-VTT, Charades, and Densevid, designed for video captioning tasks. - **Video Question Answering**: Features datasets like MovieQA and MarioQA, aimed at answering questions about videos. ### Tech Stack: The repository uses a variety of technologies and frameworks, including Python, TensorFlow, and various datasets and websites. ### Why It Matters: These datasets are crucial for researchers and developers in the field of computer vision, providing a rich set of resources for training and evaluating models on various

### Overview: This repository is a curated list of advancements in robot social navigation, focusing on dynamic and human-environmental settings. It aims to keep track of historical and recent research in the field. ### Key Features: - **Surveys**: A collection of surveys on various aspects of social robot navigation, including algorithms, challenges, and evaluation methods. - **Datasets and Benchmarks**: While not included in this repository, it lists datasets and benchmarks that are specific to social robot navigation research. - **Methods**: Covers model-based, learning-based, hybrid approaches, foundation models, and explainability, with a focus on social navigation in dynamic and human environments. - **Environment Models**: Discusses the models used to simulate social robot navigation scenarios. - **User Studies**: Reports on user studies related to social robot navigation. - **Recent Workshops**: Lists recent workshops and conferences focused on social robot navigation. ### Tech Stack: None specified. ### Why It Matters: This repository serves as a comprehensive resource for researchers, developers, and enthusiasts interested in the latest advancements in social robot navigation. It provides a structured overview of the field, making it easier to find relevant information and stay updated on the latest research.

### Overview: The dataset of Speech Recognition is a comprehensive collection of speech recognition datasets, covering various languages and applications. It includes datasets such as THCHS-30, Aishell, ST-CMDS, Primewords, aidatatang, MagicData, ASR&SD, Aishell2, TAL ASR, Common Voice, ASRU2019 ASR, 2021 SLT CSRC, WenetSpeech, KeSpeech, MagicData-RAMC, Mandarin Heavy Accent Conversational Speech Corpus, Free ST Chinese Mandarin Corpus, and more. These datasets are invaluable for researchers and developers looking to improve their speech recognition models and applications. ### Key Features: - **Chinese**: THCHS-30, Aishell, ST-CMDS, Primewords, aidatatang, MagicData, ASR&SD, Aishell2, TAL ASR, Common Voice, ASRU2019 ASR, 2021 SLT CSRC, WenetSpeech, KeSpeech, MagicData-RAMC, Mandarin Heavy Accent Conversational Speech Corpus, Free ST Chinese Mandarin Corpus. - **English**: Common Voice, LibriSpeech, ST-AEDS-20180100, TED-LIUM Release 3, Multilingual LibriSpeech, SPGISpeech, Speech Commands, 2020AESRC, GigaSpeech, The People’s Speech, Earnings-21, VoxPopuli, CMU Wilderness Multilingual Speech Dataset,

Overview: cuCIM is an open-source, GPU-accelerated image processing library designed for multidimensional images in various scientific and medical applications. Key features: - Enhanced image processing capabilities for large and n-dimensional TIFF files - Accelerated performance through GPU-based image processing and computer vision primitives - Straightforward Pythonic interface with matching OpenSlide API Tech stack: Python, Jupyter Notebook, NVIDIA CUDA Why it matters: cuCIM provides efficient and accelerated image processing solutions for biomedical, geospatial, material, and remote sensing applications, making it a valuable tool for researchers and developers working with multidimensional image data. Repository: rapidsai/cucim GitHub stars: 470 GitHub description: NVIDIA cuCIM - GPU-accelerated image processing library Tech stack: Python, Jupyter Notebook, NVIDIA CUDA Why it matters: cuCIM offers enhanced image processing capabilities, accelerated performance, and a straightforward Python interface, making it a valuable tool for multidimensional image processing in various scientific and medical applications.

### Overview: Aser is a lightweight, self-assembling AI Agent framework that simplifies the creation of intelligent agents with minimal code. ### Key Features: - **Minimal Code**: Assembles agents with just a few lines of code. - **Modular Architecture**: Provides a modular design for building agents with various features. - **Versatile Tools**: Supports a variety of tools and functionalities for different use cases. ### Tech Stack: Python ### Why It Matters: Aser simplifies the process of creating AI agents, making it accessible and efficient for developers and researchers alike.

### Overview: This repository contains the 3rd place solution for the HomeDepot Product Search Relevance competition on Kaggle. The primary language used is Python, and the repository has been starred by 469 developers. ### Key Features: - The solution includes a simplified single model, a best single model, a best ensemble model, and a best final ensemble model. - The model utilizes Word2Vec, gensim, and other libraries for text processing and feature generation. - The repository includes detailed documentation and a flowchart for better understanding of the solution. ### Tech Stack: - Python 3.5.1 - Anaconda 2.4.1 - Gensim 0.12.4 - Hyperopt 0.0.3.dev - Keras 0.3.2 - Matplotlib-venn 0.11.3 - Python-Levenshtein 0.12.0 - Regex 2.4.85 - XGBoost 0.4 ### Why It Matters: The solution demonstrates the effectiveness of ensemble models and the importance of text processing in improving model performance in a competitive setting.

Overview: Tic-Tac-Toe is a classic two-player game where players take turns marking the spaces in a 3x3 grid. The goal is to get three of your marks in a row, either horizontally, vertically, or diagonally. This project introduces the Minimax algorithm, a fundamental AI technique, to solve Tic-Tac-Toe games. Key features: - Minimax algorithm implementation for Tic-Tac-Toe - Game tree representation and game state analysis - Recursive minimax algorithm for finding optimal moves - Pseudocode and Python implementation for understanding the algorithm Tech stack: Python Why it matters: The Minimax algorithm is a powerful AI technique that can be applied to various two-player zero-sum games, including Tic-Tac-Toe. It provides a systematic way to evaluate game states and find optimal moves, making it a valuable tool in game development and artificial intelligence. Repository: Cledersonbc/tic-tac-toe-minimax Primary language: Python GitHub stars: 469 This project offers a practical example of how the Minimax algorithm can be implemented and tested, demonstrating its effectiveness in solving Tic-Tac-Toe games.

### Overview: Biomedical Information Extraction (BioIE) involves the process of extracting structured information from unstructured biomedical data. This includes text documents written in technical language, aiming to produce verifiable and consistent knowledge. BioIE has evolved significantly since the introduction of language models like BERT and more recently, Large Language Models (LLMs). Resources listed here are primarily free and have limited license requirements, focusing on methods and datasets that are publicly accessible and actively maintained. ### Key Features: - **LLMs in Biomedical IE**: Comprehensive benchmarking of sixteen different language models applied to medical language tasks. - **Ethical and Regulatory Challenges**: Review of ethical issues arising from the applications of LLMs in biomedicine. - **Pre-LLM Overview**: An overview of how BioIE and bioinformatics workflows can be applied to questions in cardiovascular health and medicine research. ### Tech Stack: None specified. ### Why It Matters: BioIE has seen significant changes since the introduction of language models, particularly with the advancements in large language models (LLMs). These models have the potential to revolutionize the field by providing more accurate and comprehensive information extraction from unstructured biomedical data.

### Overview: The Deepgram Python SDK provides a comprehensive and powerful toolset for developers working with Deepgram's speech and language AI services. This SDK simplifies the integration of Deepgram's APIs into Python applications, making it easier to build applications that leverage advanced speech and language processing capabilities. ### Key features: - **Real-time Speech Recognition (Listen v2)**: Utilizes the latest Deepgram speech recognition model with contextual turn detection, ideal for real-time applications. - **File Transcription**: Transcribes pre-recorded audio files, allowing developers to easily integrate speech recognition into their applications. - **Text-to-Speech**: Generates natural-sounding speech from text, perfect for applications that need to convert text to speech. - **Text Analysis**: Analyzes text for sentiment, topics, and intents, providing valuable insights for applications requiring natural language processing. - **Voice Agent (Conversational AI)**: Enables the creation of interactive voice agents, making it easy to build conversational applications. ### Tech stack: The Deepgram Python SDK is built using the `Fern` framework, ensuring a streamlined and efficient development experience. ### Why it matters: The Deepgram Python SDK is essential for developers looking to integrate Deepgram's advanced speech and language processing capabilities into their applications. It simplifies the integration process, making it easier to build applications that leverage the latest in speech and language AI technology.

Overview: This repository contains the code and model weights for the GILL (Generating Images with Large Language Models) model, which is designed to process arbitrarily interleaved image-and-text inputs to generate text, retrieve images, and generate novel images. Key features: - GILL is capable of generating text, retrieving images, and generating novel images. - The model is trained on the Conceptual Captions dataset and requires precomputed visual embeddings for image retrieval. Tech stack: Jupyter Notebook Why it matters: GILL is a significant contribution to the field of multimodal language models, offering a practical solution for generating images from text inputs and improving the ability of language models to generate novel content.

### Overview: The JavaScript/TypeScript SDK for Qdrant provides a comprehensive interface for interacting with the Qdrant vector search engine. This SDK simplifies the process of indexing, querying, and managing collections of vectors. ### Key Features: - **Vector Indexing**: Easily create and manage vector indexes for efficient search. - **Querying**: Perform precise and flexible queries using the provided API methods. - **Integration**: Seamless integration with both REST and gRPC protocols for different use cases. ### Tech Stack: - TypeScript - Node.js (ESM and CJS) - Deno - Browser (fetch API) - Cloudflare Workers (OpenAPI only) ### Why It Matters: The JavaScript/TypeScript SDK for Qdrant offers a powerful and efficient way to integrate vector search capabilities into various projects, making it an essential tool for developers looking to enhance their search functionalities.

Overview: This repository is a curated list of over 1,479 Python libraries and frameworks, organized by category. It serves as a valuable resource for developers looking to expand their Python skillset. Key features: - Newly Created Repositories - Agentic AI - Code Quality - Crypto and Blockchain - Data - Debugging - Diffusion Text to Image - Finance - Game Development - GIS - Graph - GUI - Jupyter - LLMs and ChatGPT - Math and Science - Machine Learning - General - Machine Learning - Deep Learning - Machine Learning - Interpretability - Machine Learning - Ops - Machine Learning - Reinforcement - Machine Learning - Time Series Tech stack: Python Why it matters: This repository is a comprehensive collection of Python libraries and frameworks, making it an essential resource for developers looking to enhance their Python skills across various domains.

### Overview: cntext is a Python library designed for researchers in computational social science who analyze text data. It offers tools for word frequency, sentiment analysis, word embeddings, and semantic projection to measure constructs such as attitudes and psychological states from Chinese text. ### Key Features: - **Word Embedding Projection Approach (WEPA)**: A method for projecting word embeddings to measure semantic similarity and distance. - **Installation**: Easy setup for integrating cntext into your projects. - **Minimal Runnable WEPA Example**: A simple example demonstrating the basic usage of WEPA. - **WEPA Documentation and Reproducibility Files**: Comprehensive documentation and reproducibility resources. - **Quick Start**: A brief guide to get started with cntext. - **Module Overview**: Detailed explanations of each module within cntext, including: - **IO Module**: Functions for reading and writing text files, PDFs, and more. - **Stats Module**: Tools for word count, readability analysis, sentiment analysis, and more. - **Plot Module**: Visualizations for text similarity and semantic projection. - **Model Module**: Implementations of word2vec, GloVe, and other embedding models. - **Mind Module**: Functions for semantic centroid analysis, concept axis generation, and semantic projection. - **LLM Module**: Built-in language models for text analysis tasks. ### Tech Stack: Python ### Why It Matters: cntext provides a comprehensive suite of tools for analyzing text data in computational social science, making it easier and more efficient to measure and understand attitudes, psychological states, and other

### Overview: AlphaEdit is a cutting-edge method that minimizes disruption to the preserved knowledge while optimizing the editing process for language models. By leveraging the null-space constrained knowledge editing technique, AlphaEdit ensures that the distribution of hidden representations within large language models remains invariant, allowing for effective knowledge update and preservation simultaneously. ### Key Features: - **Null-Space Constrained Knowledge Editing**: AlphaEdit projects parameter perturbations onto the null space of key matrices to minimize disruption to the preserved knowledge. - **Error Removal**: It removes the output error related to the edited parameters, allowing the model to focus solely on knowledge update without trade-offs. - **Mathematical Properties**: AlphaEdit utilizes the mathematical properties of matrix projection and null space to ensure invariance of hidden representations within LLMs, enabling seamless integration with post-edited models. ### Tech Stack: - Python ### Why It Matters: AlphaEdit is a significant advancement in the field of knowledge editing for language models, particularly for large-scale models like LLaMA. By optimizing the editing process and ensuring model invariance, it enhances the ability of language models to handle knowledge updates and preservation effectively, making it a valuable tool for researchers and practitioners in the field of artificial intelligence and natural language processing.

### Overview: Solon AI is a comprehensive Java AI development framework designed to integrate large language models (LLM), knowledge bases (RAG), protocols (MCP), and agent collaboration choreography. It is fully compatible with various Java frameworks such as SpringBoot, Vert.X, and Quarkus, making it highly versatile and efficient. ### Key Features: - **LLM Integration**: Solon AI seamlessly integrates with leading LLM models like OpenAI, Gemini, Claude, Ollama, DeepSeek, and Dashscope. - **RAG Knowledge Base**: Utilizes a unified interface for various RAG knowledge bases, including third-party frameworks. - **MCP Protocol Support**: Supports the transformation of agent reasoning into observable and governable computation flow graphs. - **Multi-Model Dialects**: Adapts to model differences by using a unified interface for different LLM models. ### Tech Stack: Java ### Why It Matters: Solon AI is a pivotal component of the Solon project, providing a robust framework for AI development that can be seamlessly integrated into existing Java-based ecosystems, enhancing efficiency and openness in AI application development.

### Overview: SelectionGAN is a powerful image-to-image translation model designed for cross-view image translation, achieving state-of-the-art results in the CVPR 2019 Oral and TPAMI 2022 papers. The model utilizes multi-channel attention selection and cascaded semantic guidance to enhance translation quality across different views. ### Key Features: - **Multi-Channel Attention Selection**: The model selects the most relevant channels for translation, ensuring that the final output is coherent and contextually meaningful. - **Cascaded Semantic Guidance**: The cascaded approach ensures that the semantic information is correctly aligned and propagated through the translation process. - **Cross-View Image Translation**: The model excels in translating images from one view to another, achieving high fidelity across multiple views. ### Tech Stack: - **Python**: The primary programming language used for implementing the model. - **PyTorch**: The deep learning framework used for training and inference. ### Why It Matters: SelectionGAN's ability to handle cross-view image translation with high fidelity and semantic coherence makes it a valuable tool in various applications, including panorama image synthesis and person image generation. Its performance is particularly noteworthy in achieving high-quality translations across multiple views, making it a significant contribution to the field of image-to-image translation. ### Repository: - **Repository**: Ha0Tang/SelectionGAN - **GitHub Stars**: 467 - **Primary Language**: Python - **GitHub Description**: [CVPR 2019 Oral] Multi-Channel Attention Selection GANs with Cascaded Semantic Guidance for Cross-View

### Overview: LaMDA-pytorch is an open-source pre-training implementation of Google's LaMDA research paper in PyTorch. This repository focuses on the 2B parameter implementation, aiming to provide a practical and affordable training experience for most developers. LaMDA is known for its advanced capabilities in generating human-like text, making it a valuable tool for natural language processing tasks. ### Key features: - **LaMDA Implementation**: The repository covers the pre-training architecture as detailed in the LaMDA research paper. - **RLHF Similar to ChatGPT**: Adds reinforcement learning with human feedback (RLHF) for improved quality. - **PyTorch Framework**: Utilizes PyTorch for efficient model training and inference. - **Tokenizer Integration**: Supports both Sentencepiece and GPT-2 tokenizers for better performance. - **Pipeline Parallelism**: Recommended ZeRO 1 for efficient training at scale. ### Tech stack: - PyTorch ### Why it matters: LaMDA-pytorch provides a practical and affordable way to train large language models, making it accessible to developers and researchers alike. The RLHF approach ensures that the generated text is of high quality, making it suitable for applications requiring human-like interactions. ### GitHub stars: 467 ### Repository: conceptofmind/LaMDA-rlhf-pytorch ### Primary language: Python ### About LaMDA: - **T5 Relative Positional Bias in Attention** - **Gated GELU Activation in Feed Forward Layer** - **G

Overview: This project implements a document classification model using Hierarchical Attention Networks, as described in the paper "Hierarchical Attention Networks for Document Classification" by Yang et al. (2016). Key features: - Uses Hierarchical Attention Networks for effective document classification. - Implements a TensorFlow-based solution. - Includes a pre-trained English model for spaCy for quick model loading. Tech stack: Python Why it matters: The project showcases a practical application of Hierarchical Attention Networks in document classification, demonstrating its effectiveness and potential for real-world applications.