Accessible interfaces / 2024
ASL recognition
A web app for learning and practising American Sign Language, with demonstrations and recognition feedback.
A more approachable way to practise
This web app helps people learn and practise American Sign Language. Learners watch a demonstration, try the sign on camera, and compare the word recognized by the model with the word they are practising.
My engineering focus
I worked on the web application and model-serving path: a React interface, MediaPipe landmark tracking and client-side inference. Visual overlays connect what the camera sees with the feedback shown to the person practising.
The deployment separates static assets and model weights using AWS S3. The application also includes a recording flow to support practice and data collection.
The connection
The project connects a trained model with a short, visible feedback loop: watch a demonstration, practise on camera, and see the predicted sign. It combines interface work with the practical details of running inference in a browser.
Original project screenshot
