How do you balance relevance and diversity when retrieving document chunks for RAG? — AI/ML Interview Question