LLMQuoter: Enhancing RAG Capabilities Through Efficient Quote Extraction From Large Contexts