This talk will review the engineering considerations to support the operation of body scanning machine learning models to evaluate the usability of images for generating a body double in the Amazon shopping application. In order to create next generation shopping experiences which reflect a customer’s body proportions based on their photos, the engineering team needed a low latency way to validate if the data would be able to produce a highly accurate result. The body estimation models were used to generate a 3D simulation of a custom shirt draped on the person using the iOS or Android shopping app. Engineering requirements for operating this model at the edge included latency of application download, model run time, model size, and the ability to instrument the experience for defection detection while keeping the customer data secure.