Creating a Q&A chatbot using fine-tunning with RAG

Hi everyone,

I want to create a chatbot that answers questions about customer support, I have 30000 Q&A datasets, so I guess I should use the embedding to insert our knowledge into the GPT model.
Also, I need the answer’s output language to be in a specific language and tone and prevent answering irrelevant queries so I guess I should use fine-tuning too. Still, the question is should I fine-tune the model with all of the 30000 data or just a sample of it?
Another question is if I had to add the system prompt into all new queries, is it necessary to fine-tune, or is the embedding sufficient?

nobody can help me with this problem?

I would look at embeddings for knowledge, and fine-tunes for tone. You may even be able to prompt the correct tone, but this is somewhat situationally dependent.

There’s an example in the OpenAI cookbook I would take a look at if I where you:

Hope that helps :heart: