How would you architect an MLOps pipeline for a computer vision model that needs to be deployed across thousands of edge devices?
