1. What does 'prompt versioning' solve in production LLM systems?
2. What is the primary purpose of LLM observability tooling?
3. How does semantic caching differ from exact-match caching for LLM responses?
4. What is a 'canary deployment' in the context of releasing a new LLM version?
5. Which metric best indicates that an LLM response quality has silently degraded?
6. What is 'prompt injection' from an LLMOps security standpoint?
7. Why is logging token counts per request important in LLMOps?
8. What is a model gateway in an LLMOps architecture?