I think a first step for these big enterprise codebases (also applicable to documentation) is to collect it into a big ball and finetune on it.