Reading the comments on the linked page and thinking about it a bit, one thing that is definitely impossible is having the speaker in the ball. That’d disrupt the pickup of the needle so much you’d hear a lot of noise in the background. It WOULD be possible to have the needle-ball output a bluetooth signal, not difficult at all really.

I’m no magnetophysicist but I think with the right ‘base’ underneath the LP and some good, trickly, realtime computation you could A. account for the weight of the ball and adjust for it and furthermore, B. tilt the record so that the ball slides around it after some initial motion and doesn’t skip grooves. This might make it even more sensitive to skipping though. And it wouldn’t be floating that high.