Commenting so I can find this thread easily later... as I'm also curious :).
I would also ask if you have a rough idea how much of the ML work is implementing basic models to production versus actual data scientific work in coming up with/tuning the underlying models themselves.