Once a machine learning model is trained and validated, it often feels like a major milestone has been achieved.  In reality, it’s more like the first lap in a relay race.  Deploying ML to production bears many similarities to a typical software release process, but brings several novel challenges like failing to generalize as expected