Try out a local language model for the docs you can't get chatgpt to write.

You can run small quantized models on apple silicon if you have it.

I've been using a 70B local model for things like this and it works well