Documents as parametric memory

Hello everyone,
I have a set of docs and I want to perform retrieval. So far, it’s not going great as I have to do unsupervised retrieval. I was working on a RAG model and kind of felt that is there a way I could incorporate all my docs as parametric memory like gpt without explicitly having a QnA dataset ? So I can ask a question about the docs and it can generate relevant answers.

Thanks