At what stage is the LLM included in the RAG pipeline?

In the RAG (Retrieval-Augmented Generation) pipeline, the LLM (Large Language Model) is activated during the answer generation stage. First, there is a retrieval phase where relevant information is extracted from a knowledge base or documents, and then this information is fed into the LLM, which formulates a coherent and meaningful response based on it. Thus, the LLM operates after the retrieval phase, enriching the result with context and forming the final text.