Optimizing and Automating ML Model Training Pipeline for a Leading Auto Glass Service Provider

THE BACKGROUND

The proliferation of modern technologies like ADAS (Advanced Driver Assistance System) and ability to customize vehicles makes selection of glass parts for repair and replacement a non-trivial problem. Soothsayer Analytics, in collaboration with the auto glass service provider had set up a platform called VCDX (Vehicle Characteristic Data Extrapolation) which predicted the most likely part to use for a given Vehicle Identification Number (VIN) during repair/replacement services. It had a prediction API that was deployed with a fully serverless architecture. This API was a collection of roughly 4000 models--with one model per make & model combination.