sounds close to data engineering where you shape the data you feed. Only now we’re doing it after training rather than before. Though in the context of continuously trained models the distinction seems to disappear.