Projects

### Overview: EdgeChains.js is a full-stack GenAI library that manages all core prompts and chains declaratively in JSONnet. ### Key features: - **One Script File**: Execute production-ready GenAI applications with a single script file and one JSONnet file. - **Versioning for Prompts**: Prompts are written in JSONnet, making them easily versionable and diffable. - **Automatic Parallelism**: Automatically parallelizes LLM chains and chain-of-thought tasks across CPUs, GPUs, and TPUs using the WebAssembly runtime. - **Fault Tolerance**: Designed to be fault-tolerant, with retry and backoff mechanisms even in the event of system failures. - **Scalability**: Built to be scalable, capable of writing chain-of-thought applications on large numbers of APIs, prompt lengths, and vector datasets. ### Tech stack: EdgeChains.js is built on top of JSONnet, a library originally developed by Google for managing configuration code in the Borg infrastructure. ### Why it matters: EdgeChains.js simplifies the process of deploying and managing Generative AI applications by providing a single script file and one JSONnet file for executing production-ready applications. It also handles prompt versioning and automatic parallelism, ensuring fault tolerance and scalability.

Overview: This is an unofficial Web Integration for Dialogflow V2, providing a powerful natural language understanding engine to process and understand natural language input. Key features: - Progressive Web App (100/100 Lighthouse score) - Accessibility Features - Extensive Browser Support (IE8+), offline capabilities (history) and great SEO - Familiar UI & UX, based on the official Google Assistant Design Specifications - Dark Mode & [Theming](#theming) - Hands-free interaction with Voice Input and Speech Feedback (with SSML) - Language independent - [Docker](./Dockerfile) and [Kubernetes](./k8s) support - Rich-component, Webhook and Actions on Google Support ([demo](https://mishushakov.github.io/dialogflow-web-v2/)) - Floating Widget for embedding on websites ([repo and demo](https://github.com/mishushakov/floating-chat)) - Based on Vue **3**, Webpack, Babel and PostCSS - Lightweight (build is ~50KB gzipped) - Free and fully Documented - Made in Germany Tech stack: Vue 3 Why it matters: This integration allows developers to easily integrate Dialogflow's powerful natural language processing capabilities into their web applications, enhancing user interaction and providing a rich, accessible, and SEO-friendly interface. Repository: mishushakov/dialogflow-web-v2

Overview: Redis Vector Library (RedisVL) is a Python client designed specifically for AI applications, leveraging Redis for AI-native capabilities. It offers lightning-fast vector search and enterprise-grade reliability, making it ideal for building RAG pipelines, AI agents, and recommendation systems. Key features: - Lightning-fast vector search - AI extensions for semantic caching and LLM memory management - Dev utilities for command-line interface index management - Complex filtering capabilities - Hybrid search for combining semantic and full-text signals - Rerankers for improved search relevancy - MCP server for exposing Redis indexes to MCP clients Tech stack: Python Why it matters: RedisVL provides a production-ready solution for AI applications, combining the speed of vector search with the reliability of Redis, making it a powerful tool for building intelligent systems. Repository: redis/redis-vl-python GitHub stars: 432

### Overview: Waldo is an advanced AI-powered cheat detection system designed to analyze Counter-Strike 2 gameplay footage. Utilizing cutting-edge Vision Transformer technology, it automatically extracts 2-second clips around killshots and provides detailed confidence scores ranging from 0.0 (likely legitimate) to 1.0 (likely cheating). ### Key Features: - **Analyzes CS2 Gameplay Videos**: Automatically detects killshots using audio analysis. - **Focuses on Killshots**: Extracts 2-second clips around each detected kill. - **Trains Locally from Your Footage**: Select processed clips folder and train or fine-tune a model with labels (cheating) or (not cheating). - **Analyzes Processed Clips**: Provides detailed confidence scores from 0.0 (likely legitimate) to 1.0 (likely cheating). - **Frame-by-Frame Analysis**: See exactly which 16 frames the model analyzed for each clip. - **Persistent Results**: All analysis results are automatically saved and can be revisited later. - **Export Capabilities**: Export results as JSON for further analysis. ### Tech Stack: - Python ### Why It Matters: Waldo is crucial for ensuring fair play in Counter-Strike 2, providing a reliable method to detect and prevent cheating behavior in gameplay footage.

Overview: OpenAssistantGPT is an open-source platform for building chatbot assistants using OpenAI's Assistant API, offering easy integration, low cost, and an open-source codebase on GitHub. Key features: - Easy website integration - Low cost - Open-source codebase on GitHub Tech stack: TypeScript Why it matters: OpenAssistantGPT is user-friendly and cost-effective, appealing to those seeking to integrate AI chatbot functionalities into their websites. Repository: OpenAssistantGPT/OpenAssistantGPT GitHub stars: 431 GitHub description: A Community Open-Source Saas for Crafting/Building/Creating Chatbots with OpenAI's Assistant API that you can add to your website. README excerpt: # Welcome to OpenAssistantGPT <p> <img alt="Vercel Build Status" src="https://vercelbadge.vercel.app/api/marcolivierbouch/OpenAssistantGPT" /> <img alt="GitHub Last Commit" src="https://img.shields.io/github/last-commit/marcolivierbouch/OpenAssistantGPT" /> <img alt="" src="https://img.shields.io/github/repo-size/marcolivierbouch/OpenAssistantGPT" /> <img alt="GitHub Issues" src="https://img.shields.io/github/issues/marcolivierbouch/OpenAssistantGPT" /> <img alt="GitHub Pull Requests" src="https://img.shields.io/github/issues-pr/marcolivierbouch/OpenAssistantGPT" /> </p> OpenAssistantGPT is an open

### Overview: The UCF SST CitySim Dataset is a comprehensive dataset designed to facilitate safety-oriented research and digital twin applications. It includes vehicle trajectories extracted from 1140 minutes of drone videos recorded at 12 different locations, covering various road geometries such as freeways, expressways, and intersections. The dataset also provides vehicle rotated bounding box information, enhancing safety evaluation. Compared to other video-based trajectory datasets, it offers significantly more critical safety events with higher severity, making it an invaluable resource for researchers. ### Key features: - **Vehicle Trajectories**: 1140 minutes of drone videos recorded at 12 different locations. - **Road Geometries**: Covers freeway basic segments, weaving segments, expressway merge/diverge segments, signalized intersections, stop-controlled intersections, and intersections without sign/signal control. - **Safety Events**: Includes cut-in, merge, and diverge events, with a focus on critical safety events. - **Digital Twin Assets**: Provides relevant assets like 3D base maps and signal timings for digital twin applications. - **Rotated Bounding Boxes**: Vehicle rotated bounding box information for improved safety evaluation. ### Tech stack: Python ### Why it matters: The CitySim Dataset is crucial for safety-oriented research and applications, providing high-quality vehicle trajectory data that is essential for developing and testing safety-based technologies. Its comprehensive coverage of road geometries and inclusion of critical safety events make it an indispensable resource for researchers and developers in the field of autonomous vehicles and digital twin applications.

### Overview: YOLO Vision Studio is a free, open-source Streamlit app built on Ultralytics YOLO26, YOLOE, YOLO World v2, and RT-DETR. It provides detection, segmentation, pose estimation, and tracking functionalities, including honest counts and tracking IDs. ### Key features: - **Detection**: Recognizes 80 COCO object types (YOLO26 nano → xlarge, or RT-DETR). - **Segmentation**: Provides pixel masks for detected objects. - **Pose Estimation**: Tracks 17 body keypoints and the skeleton. - **YOLO World v2**: Allows inputting text prompts to find specific objects. - **YOLOE**: Outputs boxes and masks for category names. - **Tracking**: Offers four trackers and appearance ReID, with IDs and churn counts. ### Tech stack: Python ### Why it matters: YOLO Vision Studio provides accurate and honest counts, distinguishing between different people and objects, and tracks IDs to avoid counting the same person multiple times. This makes it useful for applications where precise and reliable counting is essential, such as security monitoring, traffic management, and more.

### Overview: This repository, **Awesome-LLM-Tabular**, is a curated list of research papers that explore the integration of Large Language Models (LLMs) with tabular data. It aims to provide valuable insights and resources for developers and researchers interested in leveraging LLMs for tabular data applications. ### Key Features: - **LLM Integration with Tabular Data**: This repository focuses on papers that demonstrate how LLMs can be effectively used to process and analyze tabular data. - **Curated List of Papers**: It includes a comprehensive collection of papers that highlight advancements in tabular data processing using LLMs. - **Efficient Research**: By providing these papers, it helps save time and boost research efficiency, making it easier for developers to stay updated with the latest developments in the field. ### Tech Stack: - **LLMs**: The focus is on Large Language Models, specifically those that can be applied to tabular data. ### Why It Matters: - **Tabular Data Integration**: The integration of LLMs with tabular data is crucial in various applications, including data verification, text generation, and semantic parsing. - **Efficiency and Time Savings**: By providing curated papers, this repository helps developers save time and boost research efficiency, making it an invaluable resource for anyone working with tabular data. ### Repository: - **Repository**: johnnyhwu/Awesome-LLM-Tabular - **GitHub Stars**: 431 - **GitHub Description**: Awesome-LLM-Tabular: a curated list of Large Language Model applied

### Overview: LaminDB is an open-source data management tool designed for traceable, multimodal AI. It captures context about work, data flow, and changes, making it essential for teams in fields like life sciences. ### Key Features: - **Lineage**: Tracks results across agent sessions, notebooks, scripts, and workflows. - **Lakehouse**: Manages datasets in any format (parquet, zarr) with time travel, schema evolution, and ACID guarantees. - **LIMS & ELN**: Unified schema-based records management with support for ontologies and notes. - **FAIR Datasets**: Validates and annotates files, DataFrames, AnnData, SpatialData, etc. - **Governing Changes**: Manages changes via branching and versioning. ### Tech Stack: Python ### Why It Matters: - **Untraceable Results Cannot Be Trusted**: Especially in life sciences, hard-to-verify tasks are abundant, data formats are heterogeneous, and teams need end-to-end traceability for GxP compliance. - **Without Effective Access to Multimodal Data**: Models burn tokens or fail entirely. - **Without Governing Changes**: It's hard to evaluate agents, debug their mistakes, and safely merge their contributions. ### Repository: laminlabs/lamindb ### GitHub Stars: 426 ### Why? 1. Untraceable results cannot be trusted, especially when non-verifiable tasks are delegated to agents. 2. Without effective access to multimodal data, models burn tokens or fail entirely. 3. Without governing changes to data

### Overview: This project serves as a comprehensive guide to PyTorch, covering various aspects such as neural network layers, optimization techniques, loss functions, and more. It includes practical examples and sample codes, making it ideal for developers looking to enhance their PyTorch skills. ### Key Features: - Pytorch Playground - Model (Neural Network Layers) - Optimizer - Loss Functions - Pooling Layers - Non-linear activation functions - Basic 2 Layer NN ### Tech Stack: Jupyter Notebook - Primary language: Python ### Why it Matters: This repository provides a practical and accessible way to learn PyTorch, ideal for developers who want to enhance their skills in deep learning and machine learning. ### Repository: omerbsezer/Fast-Pytorch - GitHub stars: 430 - GitHub description: Pytorch Tutorial, Pytorch with Google Colab, Pytorch Implementations: CNN, RNN, DCGAN, Transfer Learning, Chatbot, Pytorch Sample Codes ### Running in Colab: - Clone or download all repo, then upload your drive root file ('/drive/'), open .ipynb files with 'Colaboratory' application - Download "Github2Drive.ipynb" and copy your drive root file, open with 'Colaboratory' and run 3 cells one by one, hence repo is cloned to your drive file. ([Pytorch with Google Colab](#pytorchcolab)) ### Table of Contents: - :fire:[Fast Pytorch Tutorial](#py

### Overview: Chat2Graph is a Graph Native Agentic System designed to provide a robust and scalable platform for building intelligent agent-based applications. It leverages the power of graph databases and AI to enable seamless interactions and decision-making across various domains. ### Key Features: - **Graph Native Agentic System**: Built on top of graph databases, Chat2Graph offers a native approach to agent-based systems, ensuring efficient and scalable operations. - **Dual-LLM Reasoning Machine**: Combines fast and slow thinking to enhance reasoning capabilities, making Chat2Graph highly versatile. - **Chain of Agents (CoA) Oriented Task Decomposition**: Facilitates task decomposition and graph planning, optimizing the workflow of agents. - **Structured Agent Role Management**: Ensures that agents can perform specific roles effectively within the system. - **Agent Task Compiler**: Automatically compiles tasks into executable code, streamlining the process of task execution. ### Tech Stack: - **Python**: The primary programming language used for development and implementation. ### Why It Matters: Chat2Graph represents a significant advancement in the field of graph-native agentic systems, offering a practical solution for integrating AI with graph databases. By providing a comprehensive toolkit and ecosystem, it empowers developers to build intelligent agent-based applications with ease and efficiency.

Overview: CRAB is a Python-centric framework designed to build LLM agent benchmark environments across various platforms and environments. Key features: - Cross-platform and multi-environment support - Easy-to-use configuration - Novel benchmarking suite Tech stack: Python Why it matters: CRAB simplifies the process of creating and benchmarking LLM agent environments, making it accessible and versatile for researchers and developers.

Overview: Interpret-Text is a library that incorporates state-of-the-art explainers for text-based machine learning models and visualizes the results with a built-in dashboard. It supports various NLP scenarios and provides a common API across integrated libraries. Key features: - Incorporates community-developed interpretability techniques for text models - Supports multiple explainers - Provides an interactive visualization dashboard Tech stack: Python Why it matters: This toolkit simplifies the process of building and deploying interpretable machine learning models, making it easier for developers and data scientists to experiment with different interpretability techniques and explain their models effectively.

### Overview: Data Science Hacks is a collection of tips, tricks, and hacks designed to enhance your data science skills. Whether you're a beginner or an advanced professional, these hacks will help you become more efficient and effective in your data science projects. ### Key Features: - **Resource Downloader Extension**: Quickly extract image data directly from Chrome. - **Pandas Apply**: A powerful function for manipulating data frames. - **Pandas Boolean Indexing**: Select subsets of data based on specific conditions. - **Pandas Pivot Table**: Create Excel-style spreadsheet summaries. - **Pandas Crosstab**: Visualize data in a tabular format. - **Pandas str.split**: Extract first and last names from a column. - **Extract E-mails from Text**: Easily extract email addresses from text. - **Normal Distribution**: Transform non-normal data into a normal distribution. - **Remove Emojis from Text**: Cleanse text data of unwanted characters. - **Elbow Method for Clustering**: A technique for determining the optimal number of clusters. ### Tech Stack: Jupyter Notebook ### Why it Matters: These hacks are designed to help you become a better data scientist by providing practical solutions to common problems. Whether you're working on a project, preparing data, or improving your understanding of data science concepts, these hacks will help you streamline your workflow and make your data science projects more efficient.

### Overview: gMLP is an all-MLP replacement for Transformers implemented in Pytorch, designed for masked language modeling and image classification tasks. ### Key Features: - **gMLP Implementation**: A Pytorch implementation of gMLP, an all-MLP replacement for Transformers. - **Masked Language Modeling**: Supports masked language modeling with a sequence length of 256 and 20,000 tokens. - **Image Classification**: Supports image classification with an image size of 256 and 1000 classes. - **Attention Option**: Adds a tiny amount of attention (one-headed) with an additional `attn_dim` parameter. - **Non-Square Images**: Supports non-square images and patch sizes with dimensions (256, 128). - **Multi-Head Attention**: Allows for a multi-headed approach with `heads` greater than 1. ### Tech Stack: Pytorch ### Why it Matters: gMLP is a significant contribution to the field of transformer-based models, offering an alternative to Transformers while maintaining the flexibility and performance of MLPs. Its implementation in Pytorch makes it accessible and easy to use for researchers and practitioners alike. ### GitHub Stars: 430 ### Repository: lucidrains/g-mlp-pytorch ### Primary Language: Python ### Experimental: An independent researcher proposes using a multi-headed approach for gMLPs, which can be achieved by setting `heads` to be greater than 1. ### Citations: - **@misc{li

### Overview: LVIS (Large Vocabulary Instance Segmentation) is a comprehensive dataset designed for large-scale instance segmentation tasks. It features over 2 million high-quality instance segmentation masks for 1200 object categories across 164,000 images, making it an essential resource for researchers and practitioners in computer vision. ### Key Features: - **High-Quality Segmentation Masks**: Over 2 million instance segmentation masks for 1200 object categories. - **Large Dataset**: Includes 159,623 annotated images (100k train, 20k val, 20k test-dev, 20k test-challenge). - **Annotation Files**: Provides access to annotation files for easy visualization and evaluation. - **Visualization Tools**: Facilitates the visualization of annotations and results for better understanding and analysis. ### Tech Stack: Python ### Why It Matters: LVIS is a pivotal resource for large-scale instance segmentation tasks, providing a rich dataset and tools for researchers and practitioners to advance their research in computer vision. Its comprehensive nature and extensive annotation coverage make it a cornerstone in the field. ### Repository: [lvis-dataset/lvis-api](https://github.com/lvis-dataset/lvis-api) ### GitHub Stars: 430 ### Primary Language: Python ### GitHub Description: Python API for LVIS Dataset

Overview: Automatically annotate papers using Large Language Models (LLMs). Key features: - Automatically summarizes papers - Finds and highlights key concepts and sections - Provides context as readers read Tech stack: Python Why it matters: This project helps readers understand papers more easily by providing context and highlighting important information, making it particularly useful for fields like medicine and science where detailed annotations are crucial.

Overview: This repository is a collection of awesome papers on the alignment of diffusion models, focusing on recent advancements in the field. Key features: - Alignment of Diffusion Models - RLHF/RLAIF - Directly Fine-Tuning Diffusion Models on Differentiable Rewards - Proximal Reward Difference Prediction for Large-Scale Reward Finetuning of Diffusion Models - Feedback Efficient Online Fine-Tuning of Diffusion Models - Understanding Reinforcement Learning-Based Fine-Tuning of Diffusion Models: A Tutorial and Review - Aligning Few-Step Diffusion Models with Dense Reward Difference Learning - Training Diffusion Models with Reinforcement Learning Against Sparse Rewards Tech stack: Python, Jupyter Notebook Why it matters: This collection aims to provide a comprehensive overview of the latest research in the alignment of diffusion models, making it valuable for researchers and practitioners interested in this topic.

Overview: This repository contains code implementations and experiments for replicating results from influential machine learning and deep learning research papers, focusing on classic and state-of-the-art AI/ML techniques. Each subfolder corresponds to a specific paper, model, or technique, with code, notes, and sometimes pretrained weights or results. Key features: - Implementations and experiments with various attention mechanisms - Replication and exploration of the BERT model - Conditional Generative Adversarial Networks (CGANs) - Contrastive Language-Audio Pretraining (CLAP) - CLIP and related vision-language models - Cycle-consistent GANs for image translation (CycleGANs) - Deep Convolutional GANs (DCGANs) - Distributed Data Parallel training experiments (DDP) - DeepSeek model replications and experiments - Direct Preference Optimization (DPO) and related RLHF methods - Gated Recurrent Unit models (GRU) - Long Short-Term Memory models (LSTM) - Mixture-of-Experts Transformer models (Mixtral) - Online RLHF Preference Optimization (ORPO) - Large Language and Vision Assistant models (Llava) - Low-Rank Adaptation for efficient fine-tuning (LoRA) - Moonshine model experiments - Sequence-to-sequence models (Seq2Seq) - Sigmoid Loss for Language-Image Pretraining (SigLip) - Text-to-Speech models (TTS) - Variational Autoencoders (VAE) - Vision Transformer models (ViT) - Wasserstein GANs (

### Overview: AraVec is a pre-trained distributed word representation (word embedding) project aimed at providing free and powerful Arabic NLP models to the research community. It includes six different models based on three content domains: Tweets and Wikipedia. ### Key Features: - **Models**: Provides six different word embedding models for Arabic NLP. - **Domains**: Models are built on top of Tweets and Wikipedia content. - **N-grams Models**: Offers unigrams and n-grams models with statistical techniques for generating common n-grams. ### Tech Stack: Python with the gensim library. ### Why it Matters: AraVec aims to democratize access to powerful Arabic NLP models, making them available for research and development without the need for expensive licenses. It simplifies the process of creating and using word embeddings, which are crucial for tasks like text classification, sentiment analysis, and more. ### Repository: The repository is hosted on GitHub and has been starred by 429 developers, indicating its popularity and utility in the NLP community. ### Primary Language: The project is written in Jupyter Notebook, making it ideal for interactive data analysis and model experimentation. ### GitHub Stars: 429 ### GitHub Description: AraVec is a pre-trained distributed word representation (word embedding) project which aims to provide the Arabic NLP research community with free to use and powerful word embedding models. The first version of AraVec provides six different word embedding models built on top of three different Arabic content domains; Tweets and Wikipedia. ### How To Use: -