Aggregate, process, clean, label, and version data, write and debug model code, provision compute (somehow find compute), run experiments, review results, deploy model, monitor predictions and close data flywheel loop.