Create ChatBot learned from hundreds documents with embeddings


i have question about embedding GPT-3 with hundreds documents.
I need create chatbot for thousends peoples with knowledge based on hundreds documents.
How learn model with documents and how prompting?
Firstly i upload documents via API and then create prompt with most similar embed?
Or i create vector database with chunks created from documents and find most similar guestion chunk and then i create prompt from this chunks?

Thanks for response :slight_smile:

Read this topic: