Did the ML service perform data enrichment from external sources and run the model internally, or were these functions divided between services?