Hi Sara — great article. Do you have any pointers on two aspects of multi-input Keras models that seem to be problematic (a) creating a pipeline for data prep steps — e.g. is it possible to use scikit learn pipeline to encapsulate a multi-input Keras model? (b) examples of deploying a multi-input Keras model. For example, AWS Sagemaker examples of deploying Keras models assume single input and trip on multi-input Keras models. Thanks!