In inference_pipeline line 106 we have the response generation for openai, they use a simple system prompt.
