Knowledge Share

RAG Series (13): Query Optimization — Asking Better Questions

Part 13 of the RAG series. Why does vector retrieval swing wildly based on phrasing? How does Multi-Query use multiple angles to widen recall? Why does HyDE search with a "fake answer" instead of the question? How does Query Decomposition break down complex questions? RAGAS results across four strategies: context_recall improves from 0.625 to 0.875. Full code included.