How does Data Pump use parallel during import

If you specify a parallel degree on your Data Pump import job both loading for data and metadata happens in parallel. How does it work underneath the hood?
Get the slides and watch the full recording here: dohdatabase.com/webinars/