I think I'd rather run this as an API hosted on AWS rather than locally.
When will someone cram this into a lamda with the models hosted on S3 ?