So if I take this approach then I actually would not need a knowledge base for this as I'm feeding the context in the LLM prompt, is that correct?