Projects

### Overview: This repository contains a comprehensive list of papers related to Large Language Models, covering a wide range of topics and venues. The papers are categorized by their publication dates and venues such as arXiv, NeurIPS, ICML, ICLR, ACL, EMNLP, AAAI, IJCAI, KDD, CVPR, ICCV, ECCV, IEEE, ACM, Springer, ScienceDirect, Nature, and others. The papers are updated regularly, providing a valuable resource for researchers and developers interested in the latest advancements in the field. ### Key Features: - **Coverage**: Papers from 2016 to present. - **Sources**: arXiv, NeurIPS, ICML, ICLR, ACL, EMNLP, AAAI, IJCAI, KDD, CVPR, ICCV, ECCV, IEEE, ACM, Springer, ScienceDirect, Nature, and other top AI/ML venues. - **Updates**: Automated collection of new publications. - **Features**: Advanced search, code availability tracking, and multi-venue coverage. ### Tech Stack: The repository uses Markdown and GitHub Pages to organize and present the papers. ### Why It Matters: This repository serves as a valuable resource for researchers and developers interested in the latest advancements in Large Language Models. It provides a comprehensive overview of papers from various top venues, making it easier to find relevant research and stay updated in the field. ### GitHub Stars: The repository has been starred by 16 developers, indicating its popularity and usefulness in the community. ### Related Topics

Overview: Maaagic UI is an open-source UI framework designed to empower developers with seamless integration and advanced features of AI applications. Key features: - Seamless integration with Next.js for server-side rendering - Tauri for integrating GUI for Large Language Models and stable diffusion applications - TypeScript for type safety and modern JavaScript features - Customizable components and styles - Support for various large language models Tech stack: TypeScript Why it matters: Maaagic UI simplifies the development process by providing a unified framework for building modern and feature-rich applications, leveraging the strengths of Next.js, Tauri, and TypeScript for enhanced performance and flexibility. Repository: muhammad-fiaz/MaaagicUI Primary language: TypeScript GitHub stars: 16 GitHub description: Maaagic UI is an open-source UI framework designed to empower developers with seamless integration and advanced features of AI applications.

Overview: This project is a pure Java implementation of the Qwen 3.5 inference engine, designed to be fast, lightweight, and compatible with various model families. Key features: - Single-file, dependency-free architecture based on llama3.java - Support for all tested Qwen 3.5 model families: 0.8B, 2B, 4B, 9B, 27B, and 35B-A3B (MoE) - Fast GGUF format parser - Supported dtypes/quantizations: F16, BF16, F32, Q4_0, Q4_1, Q4_K, Q5_K, Q6_K, Q8_0 - Matrix-vector kernels using Java's Vector API - CLI with --chat and --prompt modes - GraalVM Native Image support - AOT model preloading for instant time-to-first-token Tech stack: Java Why it matters: This project provides a fast, efficient, and lightweight solution for Qwen 3.5 inference, making it ideal for applications requiring high performance and minimal resource usage. Repository: mukel/qwen35.java Primary language: Java GitHub stars: 16

Overview: DataWhisperer is an NLP text to visualization builder for Tableau Desktop, designed to help users create interactive visualizations from their text data. Key features: - Automatically saves and loads workbook files - Supports Python 2.7 and Tableau Professional Edition - Allows querying of text data for visualizations - Provides a graphical interface for building and customizing visualizations Tech stack: Python 2.7, xml, tkFileDialog, Tableau Professional Edition Why it matters: DataWhisperer simplifies the process of turning text data into interactive visualizations, making it easier for Tableau users to explore and analyze their data.

Overview: A local, zero-config Retrieval-Augmented Generation (RAG) system built with FastAPI, Streamlit, Qdrant, and Ollama for efficient and offline querying of documents. Key features: - Upload PDFs and extract content - Embed content using Sentence Transformers - Store and search with Qdrant vector database - Query documents using local LLMs via Ollama - Smart chunking and clustering support - Offline functionality - Fast retrieval and response generation - Semantic search with relevance scoring Tech stack: Python, Docker, FastAPI, Streamlit, Qdrant, Ollama. Why it matters: This project simplifies the setup of a local RAG system, requiring no manual installation, configuration, or model downloads, making it ideal for developers and researchers looking to quickly integrate local LLMs into their workflows without the hassle of traditional RAG setups.

### Overview: OmniMemory is a production-ready memory framework designed for AI agents, offering dual-agent synthesis, self-evolution, and composite scoring capabilities. It updates, deletes, or consolidates knowledge graph entries based on conflicts, ensuring a more dynamic and efficient memory system. ### Key Features: - **Dual-Agent Synthesis**: Two specialized agents analyze behavior and narrative to resolve conflicts and update the memory graph. - **Self-Evolving Memory**: Automatically detects and resolves conflicts between new and existing information, updating, deleting, or consolidating knowledge. - **Composite Scoring**: Returns the most useful memory based on relevance, recency, and importance, ensuring high-relevance memories always win. - **Enterprise Multi-Tenancy**: Built for SaaS, providing physical, logical, and session-level isolation for multi-tenancy. ### Tech Stack: Python ### Why It Matters: OmniMemory transforms static embeddings into a self-evolving cognitive substrate, offering a more dynamic and efficient memory system compared to traditional RAG, which only stores data.

Overview: VSTAR is a multimodal dialogue dataset for situated semantic understanding that incorporates scene and topic transitions. It includes dialogues, feature extraction, scene and topic segmentation, and dialogue generation. Key features: - Dialogues with scene and topic transitions - Feature extraction using ResNet and RCNN - Scene and topic segmentation - Dialogue generation using a pre-trained model Tech stack: Python Why it matters: VSTAR provides a valuable resource for researchers and developers interested in dialogue understanding, scene and topic segmentation, and multimodal dialogue generation, offering a comprehensive dataset for evaluating and advancing these technologies.

### Overview: This project demonstrates how to create a Next.js chatbot using Pinecone as the search backend and the OpenAI API for text generation. It showcases the use of API Routes for streaming capabilities and integrates Pinecone for efficient indexing and search. ### Key Features: - Next.js - OpenAI API (ChatGPT) - streaming - API Routes (Edge runtime) - streaming - Pinecone store ### Tech Stack: TypeScript ### Why It Matters: This example provides a practical demonstration of integrating AI with Next.js, Pinecone, and the OpenAI API, showcasing how to build a chatbot that can generate human-like responses. It is useful for developers looking to deploy AI-powered chatbots on Vercel or other cloud platforms.

### Overview: Jigen DB is a vector database written from scratch in C# for the .NET platform. ### Key features: - **In-process engine (`Jigen.Store`)**: append-only memory-mapped storage, asynchronous ingestion, crash recovery, exact (brute-force) search. - **HNSW index** (`Jigen.Indexer.HNSW`): disk-backed approximate nearest neighbor graph with concurrent inserts, deletes, optional SQ8 quantization and exact reranking. - **Sparse index** (`Jigen.Index.Sparse`): per-collection inverted index over term/weight pairs, with segment compaction and hybrid dense+sparse search. - **Server**: multi-database host with gRPC (port 3223) and REST (port 13223) APIs, per-collection dense, sparse, and hybrid search with content filters, periodic durability checkpoints. - **Embeddings**: server-side text embedding generation (ONNX), in-process or scaled out to dedicated worker containers over RabbitMQ; CPU by default, GPU execution providers available. - **Typed .NET client** (`Jigen.Client`): dictionary-like collections, dense/sparse/hybrid search, LINQ predicates translated to server-side filters. - **Python client** (`jigen-client`): sync/async gRPC API, MessagePack-compatible collections, and optional LangChain/LlamaIndex adapters. ### Tech stack: **.NET (net8.0 / net10.0)**, **ASP.NET Core**, **gRPC**, **ONNX Runtime**. ### Why it

Overview: This project aims to implement forest fire detection using a Convolutional Neural Network (CNN). It utilizes a basic pretrained model for training and testing on real-life images and videos for classification and object detection. Key features: - Implementation of a CNN for forest fire detection - Utilization of a pretrained model for training and testing - Data analysis and preprocessing - Neural network creation from scratch - Pretrained model setup - Data augmentation for extended dataset - Video fire detection - Project launcher Tech stack: Keras, Matplotlib, numpy, pandas, TensorFlow, ImageIO, Computer Vision (OpenCV). Why it matters: This project serves as the foundational version of a forest fire detection model, achieving 98% accuracy through training and testing. The project aims to reduce complexity in future versions, use the latest pretrained model for improved accuracy, and provide an end-to-end fire detection model with object detection capabilities.

### Overview: Learn fine-tuning techniques for text-to-image models and discover how to teach a diffusion model a concept of your choosing, such as a special style or a pet, with minimal image input. ### Key features: - Fine-tuning techniques for text-to-image models - Concept teaching with as little as 3 images - Minimal image input for concept teaching ### Tech stack: Jupyter Notebook ### Why it matters: Gain hands-on experience with text-to-image diffusion models and discover practical methods for personalization and subject-driven generation.

Overview: This project provides a comprehensive end-to-end deployment template for Azure OpenAI applications using Redis Enterprise as a vector database. Key features: - Azure OpenAI instance with `text-davinci-003` and `text-embedding-ada-002` models - Azure Redis Enterprise as a vector database - Azure Storage bucket for file storage - Azure Web App deployment Tech stack: Terraform Why it matters: This template streamlines the deployment process, making it easier to set up and manage Azure OpenAI applications with Redis Enterprise as a vector database, enhancing efficiency and scalability. Repository: redis-applied-ai/azure-openai-redis-deployment GitHub stars: 16 GitHub description: End-to-end deployment of Azure OpenAI applications using Redis Enterprise as a vector database.

Overview: This project is an implementation of a Retrieval-Augmented Generation (RAG) system using Llama Index and Qdrant vector stores. It allows users to index and query PDFs using a language model for enhanced text generation. Key features: - Index and query PDFs using Llama Index - Utilize Qdrant for efficient vector storage - Generate output using a language model Tech stack: Python Why it matters: This project showcases the integration of advanced indexing and retrieval techniques with natural language processing, providing a powerful tool for text generation from structured data like PDFs. Repository: romilan24/llama-index-RAG Primary language: Python GitHub stars: 16 GitHub description: A RAG implementation on Llama Index using Qdrain vector stores as storage. Take some pdfs, store them in the db, use LLM to inference.

### Overview: This project is a set of scripts designed to generate full attention-head heatmaps for transformer-based language models, specifically focusing on the Qwen model. These heatmaps help visualize how the model processes input elements, identifying key tokens and features, and understanding the model's attention patterns across different layers and heads. ### Key Features: - **Attention Head Heatmaps**: Detailed visualizations showing the model's attention weights across all layers and heads. - **Visualization Tools**: Provides tools to animate and analyze attention heatmaps for better understanding. - **Model Compatibility**: Works with various transformer-based models, including Qwen, to provide insights into their behavior. ### Tech Stack: - **Python**: Scripts are written in Jupyter Notebook. - **Libraries**: `transformers`, `torch`, `matplotlib`, `seaborn`. ### Why It Matters: Understanding how transformers process information and identify relevant tokens is crucial for debugging biases, improving prompt engineering, and designing more efficient architectures. By visualizing these patterns, researchers can gain deeper insights into model behavior and make informed decisions for future model design and optimization. ### Repository: - **Repository**: `ronniross/attention-heatmap-visualizer` - **GitHub Stars**: 16 - **GitHub Description**: A set of scripts to generate full attention-head heatmaps for transformer-based language models. ### Part I - 2026 Heatmaps: - **Visualization of Qwen 3.5 Heatmaps**: Updated static full heatmap for Qwen 3.5. - **Attention-Heatmap Animator**: Animated visualization of attention

Overview: The Symbiotic Core Library offers a framework for ethical AI principles, practical modules, and research grounded in societal and environmental considerations, addressing urgent cognitive and socio-economic challenges in AI development and deployment. Key features: - Ethical principles - Practical modules - Research grounded in societal and environmental considerations Tech stack: Node.js, TypeScript Why it matters: This library aims to foster a healthy, co-evolutionary technological ecosystem by focusing on the practical, ethical, and societal dimensions of AI development and deployment, addressing urgent challenges like cognitive degeneration and environmental impacts.

### Overview: rvQR is a mobile-first, no-network, no-account, SHA-256 verified optical transfer tool for RVF cognitive containers and WASM artifacts. It allows files to be moved between devices without requiring internet connectivity, paired accounts, or installation. ### Key Features: - **Offline Optical Transfer**: Move files between devices without needing an internet connection. - **No Setup Required**: No pairing, no accounts, no installation. - **SHA-256 Verification**: Files are verified and stored using SHA-256 hashes. - **Two-Way Transfer**: Can send and receive files between devices. - **No Camera Needed**: Works with any browser that supports camera access. - **Standalone App**: The entire app is included in a single file, making it completely self-contained. - **Real Artifacts**: Both the demo artifacts and the RVF microkernel are included in the repo. ### Tech Stack: JavaScript ### Why It Matters: rvQR simplifies the process of transferring files between devices, making it accessible and secure in remote or air-gapped environments. The tool's ability to work offline and verify file integrity makes it particularly useful in scenarios where network connectivity is unreliable or nonexistent. Repository: [ruvnet/rvQR](https://github.com/ruvnet/rvQR) GitHub Stars: 16

Overview: Embeddit is a Python/flask-based application that enables users to search for images using text queries or by uploading images. It utilizes OpenAI's CLIP for image embedding and LanceDB for efficient similarity search, making it a powerful tool for image-based and text-based image searches. Key features: - Text-based image search - Image-based search - Efficient similarity search using LanceDB - User-friendly web interface - Customizable image folder for indexing and searching Tech stack: Python/flask, OpenAI CLIP, LanceDB Why it matters: Embeddit simplifies the process of finding images based on text descriptions or visual content, making it an invaluable tool for developers and designers looking to enhance their image search capabilities.

Overview: Multitype-LLM-Chat is a prototype project that integrates various data sources such as YouTube videos, PDFs, and internet links, alongside an intuitive user interface, to enhance the research experience by augmenting GPT models with custom data. Key features: - Integrates multiple data types - Provides an intuitive user interface - Enhances research experience through augmented GPT models Tech stack: TypeScript Why it matters: This project showcases the potential of combining AI with diverse data sources to create a more comprehensive and accurate research tool, making it valuable for researchers and data enthusiasts alike.

Overview: This repository provides a step-by-step guide on fine-tuning a Faster R-CNN model for improved 2D object detection accuracy using PyTorch. The project focuses on enhancing the accuracy of object detection tasks with a custom dataset. Key features: - Fine-tuning of a Faster R-CNN model - Transfer learning from a pretrained model - Data augmentation techniques - Custom dataset preparation for vehicle detection Tech stack: Jupyter Notebook Why it matters: This guide is useful for developers and data scientists looking to improve the accuracy of object detection tasks, especially when working with specialized datasets. The repository offers a comprehensive walkthrough of the fine-tuning process, making it accessible for beginners and advanced users alike.

Overview: This project implements Retrieval Augmented Generation (RAG) using OpenAI's embedding models and LangChain's Python library, aiming to create a user-friendly application for querying and retrieving information from various sources. Key features: - Implements RAG using OpenAI's embedding models and LangChain's Python library. - Supports document splitting, embeddings, vector database (Chroma/FAISS), semantic search types, and retrieval chains. - Provides a web-based interface for easy data ingestion and retrieval. Tech stack: Python with Jupyter Notebook. Why it matters: This project offers a practical solution for developers looking to integrate RAG functionality into their applications, leveraging the power of LangChain and OpenAI's embedding models for enhanced information retrieval and generation capabilities. Repository: sienlonglim/LangChain Primary language: Jupyter Notebook GitHub stars: 16 GitHub description: This project implements RAG using OpenAI's embedding models and LangChain's Python library.