Pseudonymisation only helps you where the data sans the PII is still useful to you and you wan to retain it beyond the scope of being allowed to use the PII itself. That‘s a tricky thing though, because you can e.g. randomize the name stored for a user, but if you only got a single user in a certain village they might still be identifiable whereas location data in a big city might not come with the same risk. This - by my understanding - is a lot about statistics.
Your suggested architecture is still about having access to the PII though. Doesn‘t matter if it‘s in a process or in some database. They just come with slightly different requirements for getting access.
