Empowering in-browser deep learning inference on edge devices with just-in-time kernel optimizations.

Shows an example how to train a keras model for brain lesion segmentation and how to use keras.js to run the inference path in the browser

Cat & Dog Classifier is a browser-based machine learning application that classifies uploaded images as either cat or dog using a convolutional neural network. The project demonstrates how deep learning models can be integrated into a modern frontend stack using React, TypeScript, and TensorFlow.js, enabling real-time inference directly in the brow

MazeRunner project is an interactive, browser-based visualization of a Deep Q-Network (DQN) agent learning to solve a 10x10 maze. The entire reinforcement learning process, from model creation to training and inference, runs directly in your browser using TensorFlow.js.

Compile trained Burn deep learning models to WebAssembly for browser-based inference with optional WebGPU acceleration

A web‑native CNN workbench for training, inference, and Grad‑CAM visualization of image classification models instantly with — Deep Learning in your browser.

Web demo for deducing customer shopping behaviors using Deep Learning. Features AlexNet and ResNet-50 models exported via ONNX for browser-based inference.

A lightweight web application for abdominal CT segmentation using the SuPreM deep learning model. Built with FastAPI and Docker to give GPU-based inference through the browser. Project is a demo for the developer role in Professor Zhou's lab @ JHU

Built a deep learning image classifier that identifies 10 types of fast food from photos. Used transfer learning with ResNet50V2 pre-trained on ImageNet, fine-tuned on a balanced dataset of 15,000 images. Achieved 88% test accuracy. Model is exported to TensorFlow.js for in-browser inference with no backend required.

This repository contains a deep learning-based image classification system that detects whether an image contains an amora or not. The project includes dataset preprocessing, model training using TensorFlow/Keras, and a Next.js web frontend that runs the trained model directly in the browser using TensorFlow.js for real-time inference.

This is an automated medical image segmentation, interactive 2D/3D visualization, and AI-assisted clinical workflows. The platform supports DICOM/NIfTI data ingestion, GPU-accelerated deep learning inference (nnU-Net/U-Net) for lung, brain, and spleen segmentation, and real-time 3D rendering in the browser. It also includes RAG-based AI chatbots.