There might be an effective way, but it requires a programmed agent (rather than just a simple chat interface). You don't need to finetune a model or anything like that, but you DO need to provide enough context. Fortunately you have 24,000 words.

This is just broad strokes and is missing nuance, but here's what I would do, generally speaking:

Have an LLM extract your writing style into a set of rules + examples (e.g. a writing_style.md) - just feed it 20 pages and say "create rules" then take the output. Then create a separate file of examples of your actual writing from the book so far "samples.md". Lastly, you'll want to inform the system of your intended conclusion for the book and any major events that happen between now (where you are in your writing) and the ending. Call this events_and_conclusion.md. Once you have all 3 artifacts then you build an agent that has creative storytelling responsibility. We have gone from A-R, now we need the agent to construct the path from S-Z with certain events along the way. Tell the agent to think creatively and come up with the major things that happen (e.g. fill in the gaps of the story). Then once you have enough working pieces, have the agent string paths between each of the major events (e.g. what are the micro events that progress the story to the next stage). Lastly, use the style_guide and samples files to polish/refine the language in the story itself. So you're starting with the big pieces of completing the narrative, then you're glossing over it with stylistic context. Again, not too easy but nothing crazy either.