RAFT and reflexion in my pipeline

API
,
1

I have an rag application for our customer service and i want to fine tune the model to the domain specific knowlege. I have read an paper about RAFT .

I want to implement something similar to this but i am not sure if my approach is good or not?

  1. Collect all answers and the needed documents from our rag application
  2. To all wrong answer from our rag application, we use the correct answer from our support.
  3. Create {Q,A,D} where Q is the question is from the user, A is the correct answer from our support or chatbot and D are all retrieved documents which are used for the answer generation.

My question is: Is this methodology practical or should i do this exact like in the paper?

My intuition tells me that this is methodoly seems sufficient?