You should probably explain the difference between Deep MTL and MTL for those who don't know the difference.
A "normal" MTL is done by the translation service of Google. A translator that offers a very consistent result, but a not very "gammar loving" one, to the point that a non native English speaker has trouble figuring out what the hell is going on.
People and certain sites usually just batch-translate everything with Google-Translate and upload without checking anything, which is also why they are usually very buggy.
DeepL is a small company that was founded 2017 and is trying to use machine learning to provide a better translation result.
It somewhat succeeds in that it "can" provide translations that are much more coherent and less jarring to read.
The big problem DeepL has is that the AI is still very inconsistent.
Sometimes it completely crashes, gives out weird error messages, duplicates sentences, or even just returns the original Japanese input.
Because of that DeepL cannot be used for a fully automated MTL.
So somewhat simplified what I'm doing to try to fix that when creating a "slightly edited DeepL/Google hybrid MTL" (short DMTL) is batch-translating everything with DeepL and then checking every single sentence manually one by one looking for broken lines.
Whenever I see a broken line, I run it through DeepL again.
If it's still broken, I replace it with Google-Translate.
If the google entry is also broken, I manually translate the section.
Then I start the game and look for errors, remaining Japanese text, scripts, picture based text, etc and I try to manually translate those.
I wouldn't really call my stuff "proofread" though, because to keep the workload more reasonable, I only look for outright broken stuff, I will not try to fix grammar or moaning/slapping sound sections.
(A DMTL is not meant to replace a proper translation, it's supposed to make sure a game can be enjoyed in the west even if it's never going to get a proper one.)