Thanks, cool question. The insane performance gains we have when visualising are mostly because of conceptual decisions we made. One such thing is that we don't try to plot more datapoints than there are pixels on the users screen. But when a user does calculations on data (e.g. a low pass filter), we really have to do all the math. So in that case we are mostly using numpy (python) operations, which are optimised in C already. But I don't know if this includes vectorisation