TabNine sometimes predicts code completion far and above what pycharm can do.
However, that plugin maker is unlikely to tune it to work really well in any one of the IntelliJ tools. For example, on python it often tries to add semicolons to the end of lines. Tabnine also conflicts with python auto import on pycharm.
It seems like the idea of using a python data set from github and more predictive behavior would be hard but the payoff would be massive.
The algorithm could be tuned over time, and IntelliJ is in a great position of customer privacy for building datasets off customer products since the company's software already directly handles code.