At DimensionLab we are building a platform called SIML.ai for training “learned simulators” for Computational Fluid Dynamics leveraging deep learning. We quickly realized that one part of the product focused on training and optimizing these models could be used in a more general way for other types of models. Therefore, I’m interested in what was the challenges in setting up the infrastructure for training your models and what could make it significantly easier? What are the features that you’d find missing from the cloud-based training infrastructures available now from various companies? Thank you for your answers in advance! They will be extremely helpful for informing the feature-set and UI/UX decisions for our products.