We achieved this by simply doing prompt engineering. https://docs.llamaindex.ai/en/stable/examples/prompts/prompts_rag/