Retrieval-Augmented Generation (RAG) sounds simple on paper: embed your documents, retrieve the relevant chunks, stuff them into a prompt, let the LLM answer. It works great in a weekend demo. Then you ship it, real users start typing real questions, your knowledge base grows past a few hundred documents, and suddenly the "simple" system starts hallucinating, retrieving the wrong chunks, or timing out. This article walks through five challenges that show up almost every time RAG goes from pr...