Is there a way with llama index of

this might be a bit of prompt engineering on the query part. we already have outer instructions telling the LLM to ignore prior knowledge and only use the context

Thanks Jerry!