Tech »  Teaching the model: Designing LLM feedback loops that get smarter over time