How would you implement human-in-the-loop feedback for fine-tuning LLM outputs in production

Question

With the help of code can you tell me How would you implement human-in-the-loop feedback for fine-tuning LLM outputs in production?