The goal here is to generate a typed API across a bunch of microservices (written in some typed language suited for the job) that are consumed by a Python frontend. The current design is a pile of vertically-integrated monoliths that touch the disk, database, perform backend operations and rendering all in one process.

Python's single-threaded design makes it difficult to be responsive to small queries quickly while simultaneously serving large, time-consuming queries (i.e. git operations). You can get around this using worker queues to separate interpreter processes and an async design, or otherwise splitting your workload up... or you can use a language where "have a threadpool" is actually a properly supported concept, and an architecture where sharding git/email/etc backends is feasible.