setup real-time tensorflow inference using AWS sagemaker endpoint and AWS inferentia

We have a python script that uses a tensorflow model and keras model to extract data from an image and output the result. Currently it is using CPU with flask, however performance is bad and takes 20+ seconds to compute a single image. We require this to be imported to AWS and utilize a low cost AWS inferentia instance whilst performing the inference.

its a relatively simple task:-
https://introduction-to-inferentia.workshop.aws/inf1onsage/02sageintro.html