Great concept. Would like to see more of this idea applied to neural network processing and configuration in general (which in my experience can sometimes be a tedious, hard-coded affair).