Design an LLM chatbot with RL-trained answers

Design a chatbot that answers users' questions in real time, built on a large language model.

Cover the serving path (latency, streaming answers, scale), then the learning loop: where the training and fine-tuning data comes from, which reinforcement-learning tasks you would set up to make the answers better, how the rewards are defined, where those rewards can be gamed or mislead the model, and what you would do about it.