Review on DataFleets - Federated Learning and SQL by Mike Wallace

The ability to train and deploy models in near real-time from our own cloud infrastructure. It's not perfect, but it's improving quickly. The support for Keras is limited compared to the other frameworks I've used (Tensorflow, CNTK). They're working on that, but it would be great if they could work with TensorFlow more. The biggest issue was building out enough capacity to have models trained in near real time. We were able to solve this by using the Google Cloud ML Engine service, which is free up to certain usage limits.