[dvs13]

One thing I didn't get to in this article is whether metadata is always important for machine learning. If the approach is "one model does everything", then finding a suitable model isn't a problem (just choose your least hated cloud provider!)