Using a Feedback Loop for LLM-based Infrastructure as Code Generation