EobardThawne2/MudraAI ? reverse-engineered prompt

Build me a real time web app called Mudra AI for Bharatanatyam practice. I want someone to open it in the browser, allow webcam access, and get instant frame by frame feedback on the 28 single hand mudras, so they can see which gesture the app thinks they are doing and whether they need to correct it. The hand tracking should happen in the browser, and the prediction should come from a backend model so it feels fast and responsive.

Please make the experience clean and simple, with a home screen and a main practice studio view. Show the live camera, the detected hand landmarks or overlay, the current predicted mudra, and clear status messages for loading, camera permission, and connection issues. Make sure the frontend and backend run locally together, and keep it ready for easy deployment with the frontend on Vercel and the backend on Render. If there is already a trained model in the repo, wire it up and use that. Look up current docs online if you need to.