Guiding the LLM to talk / act on a way written on the document.

[djsteve1223]

Hi. I'm currently using langchain to develop a customer support chatbot for a e-commerce website.

Currently, I've managed to make it answer correctly based on policy or consumer data.
However, I also want it to talk and behave in a way written on a certain document.

I have a customer support manual (writtern for humans) which tells

1. What action plan you(LLM) should take (ask for order number, home address first, then search ~~ etc.)
2. And the manner in which you(LLM) should talk to the customer.

It's too long to prompt it as a whole.

I want to feed this into the LLM, but I have no clue where to start from.
The first way that came up in my mind is store the document as embeddings first, retrive the specific rules and manners it should behave when the query comes in, then add it on the prompt.

Any refernces, papers, links, suggestions, opinions would be helpful!