Question bank

4,310 interview questions, answered.

Reference answers, what the interviewer is really testing, how it is graded, and the follow-ups that come next.

Easy 7Medium 19Hard 36
Filters · onDifficulty, topic, company
Clear all

36 questions

  1. 1A/B testing LLM productsHardAI ML
  2. 2Agent memory: short-term vs long-termHardAI ML
  3. 3Catastrophic forgetting during fine-tuningHardAI ML
  4. 4Common agent failure modes in productionHardAI ML
  5. 5Designing LLM benchmarks: MMLU, GSM8K, HELMHardAI ML
  6. 6Detecting distribution shift in production LLM trafficHardAI ML
  7. 7DPO vs RLHF/PPO: when would you pick each?HardAI ML
  8. 8Evaluating reasoning models vs chat modelsHardAI ML
  9. 9Fine-tuning vs RAG vs prompt engineeringHardAI ML
  10. 10FlashAttention and why it mattersHardAI ML
  11. 11Guardrails for agents that call external APIsHardAI ML
  12. 12How do you evaluate a RAG system?HardAI ML
  13. 13How do you evaluate an agent?HardAI ML
  14. 14Knowledge distillation for LLMsHardAI ML
  15. 15Latency vs throughput vs cost: choosing batch sizesHardAI ML
  16. 16LLM-as-judge: when it works, when it failsHardAI ML
  17. 17LLM observability vs classical ML observabilityHardAI ML
  18. 18Measuring hallucination rate without expensive human evalHardAI ML
  19. 19Mitigating hallucinations in RAGHardAI ML
  20. 20Monitoring RAG in productionHardAI ML
  21. 21Multi-agent orchestration: when does it help?HardAI ML
  22. 22Paged attention (vLLM)HardAI ML
  23. 23Planning and decomposition in multi-step agent tasksHardAI ML
  24. 24Prompt caching and KV cache prefix sharingHardAI ML