Synapse Health interview question

How would you properly handle LLM hallucinations in agentic workflows?