We could have an option somewhere to re-train based on recent interactions, too