Most RAG approaches use a vectorDB and embeddings for schema linking. In this case the fine-tuning is handling schema linking and there is no vectorDB.