How would you architect a scalable training pipeline for computer vision models using specialized hardware and sensor data?
