Through model cascading, first-stage ML algorithms perform pre-filtering before more power-hungry models are executed.