the neat thing about the vector implementation is that we can immediately return the relevant pages while GPT formulates an answer, which gives this a hybrid "search + ask" function. We'll aim to do that in the next iteration