It is similar to the retrieval stage, yes. The concept of RAG is to then feed the retrieved fragments to a model so it can generate an answer that takes them into account.