Pergunta de entrevista da empresa S&P Global

How could I implement RAG so that LLM should not answer to its trained data and only able to answers regarding ingested data