Using the Question Answering with RAG Accelerator from the watsonx resource hub as an example, we explain how to build RAG applications with watsonx.ai and watsonx Discovery / Elastic Search at scale. We introduce a Python pipeline in watsonx.ai for document conversion, splitting, bulk import of chunks and vector indexing with ELSER vector embeddings in watsonx Discovery. We show how the Q&A with RAG Python function deployment end point can be invoked from applications, how all questions, retrieval results, answers, and end user feedback on answers can be collected in a log index in watsonx Discovery, and how the data in the log index can be analyzed using a Notebook in watsonx.ai to detect topics and report by topic how answers were rated by end users.