Training ML models is only the first step, when you need to deploy and operate them at scale, you need to ask yourself what are the options to execute the inference in the most efficient way