A RAG system has high context precision but low faithfulness. How would you address this? — AI/ML Interview Question