Towards Formal Verification of LLM-Generated Code from Natural Language Prompts