What is the proper way when you are building an API that will execute a sagemaker call for inference on every request received on FastAPI app?