Using an LLM for query planning in RAG –> 40% better answer relevance | Dark Hacker News