This looks very cool. Do you see it as tightly coupled to ML or would you think this is adaptable to more general tracking of dataset processing? It looks pretty flexible to me after a quick glance but I haven't had time to dig deeper and understand the details of how it works.