The app runs TensorFlow Lite with MoveNet for pose estimation from the camera feed, then classifies detected poses as good or bad posture using a trained fully connected neural network. All AI inference happens locally on the device without network connectivity, and the app provides voice alerts when poor posture is detected. The project includes training code in a Jupyter notebook for the pose classifier.