That's a fascinating use-case! I don't know how much of that is supported by NotebookLM - I guess that depends on the quantity of notes and guides - but the way I would approach it is to categorize the sources and gather them in individual NotebookLM units. That way I could then interrogate the related content, which is pretty analogous to how our brain works.

Otherwise, there are quite a few good open source LLMs available that you could train and run on your own computer, that might be a better solution.