Comment by vvipgupta

Thanks for the suggestion and links. Completely agree, ML production data management can be painful and to support model refinement for users that operate at scale, an abstraction at the data layer would be a useful feature.