Sofia is an artificially Intelligent yoga trainer flutter app which can detect yoga poses in real time. Uses tensorflow-flite for ON-Device ML. The yoga pose dataset was obtained and cleaned manually. The model used transfer learning on VGG16. The f1 score achieved was 0.98. The model is hosted online with AWS and has REST API endpoints for external access. The flutter app uses these endpoints to send snapshots to the model and get the pose as output. Uses tensorflow-flite for ON-Device ML to remove the latency issue for multiple simultaneous users.