Running into a number of edge-cases, errors and unreliability issues when using Retrieval Augmented Generation (RAG)? I look into some of the challenges and pragmatic solutions one could use.