adding your 'accurate' data sources, the problem with long books is you'll have to copy and paste (chunk) them up and keep sending them to the LLM. you can also download like ollama and run a local language model and do the same thing (I've programmed those for clients that don't want their data going to openai across the internet and not having to pay their per token api cost), but yeah you'll need the text somehow, I think GPT can do OCR on the pages if you can take pictures of them, I haven't tried, but that sounds tedious. RAG sounds fancy but all it literally is just like copy and pasting into the LLM as part of the conversation.