At the risk of sounding LLM-obsessed, this sounds like a great use-case for Claude.

Even when your input is incoherent, it’s often enough to get the AI in the right direction, and then it’s easier to edit.