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

62 questions

  1. 25Latency vs throughput vs cost: choosing batch sizesHardAI ML
  2. 26LLM-as-judge: when it works, when it failsHardAI ML
  3. 27LLM observability vs classical ML observabilityHardAI ML
  4. 28LoRA vs full fine-tuningMediumAI ML
  5. 29Measuring hallucination rate without expensive human evalHardAI ML
  6. 30Mitigating hallucinations in RAGHardAI ML
  7. 31Model Context Protocol (MCP)MediumAI ML
  8. 32Monitoring RAG in productionHardAI ML
  9. 33Multi-agent orchestration: when does it help?HardAI ML
  10. 34Overfitting: detection and preventionEasyAI ML
  11. 35Paged attention (vLLM)HardAI ML
  12. 36Planning and decomposition in multi-step agent tasksHardAI ML
  13. 37Precision vs recall vs F1 vs AUCEasyAI ML
  14. 38Prompt caching and KV cache prefix sharingHardAI ML
  15. 39QLoRA vs LoRAHardAI ML
  16. 40Quantization: INT8, INT4, FP8, AWQ, GPTQHardAI ML
  17. 41RAGAS: why it's popular and where it breaksMediumAI ML
  18. 42Regression testing before model/prompt changes shipHardAI ML
  19. 43Retrieval over structured data: tables, code, JSONHardAI ML
  20. 44RLHF in plain terms: what does it solve that SFT cannot?HardAI ML
  21. 45Scaling LLM inference for traffic spikesHardAI ML
  22. 46Sinusoidal positional encodings vs RoPEHardAI ML
  23. 47Speculative decoding: when it helps, when it failsHardAI ML
  24. 48The lineage from PPO to DPO to GRPOHardAI ML