Retrieval-Augmented Generation (RAG) has become the standard architecture for LLM applications that need accurate, up-to-date information. However, naive RAG implementations often fail...