Can this be used in conjunction with a text-generating LLM for better quality code generation?