What would the product be? My experience is that most ML papers are very brittle, for every cool result/example you see there's a plethora of nonsense spit by the models.