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 190Medium 2,498Hard 982
FiltersDifficulty, topic, company

4,310 questions

  1. 289How would you build an evaluation suite for vision-language models?Hard· 20 min
  2. 290How would you build an evaluation suite for voice cloning?Hard· 20 min
  3. 291How would you capacity-plan an inference service with bursty traffic?HardSystem design· 30 min
  4. 292How would you choose between an open-source and proprietary model?MediumAI ML· 10 min
  5. 293How would you debug an AI deployment using a customer’s private data?HardTechnicalCoding· 30 min
  6. 294How would you decide between prompting, RAG, fine-tuning, and training a model?Hard· 20 min
  7. 295How would you defend an LLM application against prompt injection?HardTechnicalSystem design· 30 min
  8. 296How would you define success metrics for a generative-AI product?HardProductData· 20 min
  9. 297How would you design an AI assistant for doctors?HardProduct· 30 min
  10. 298How would you design multi-tenant retrieval without leaking customer data?HardTechnicalSystem design· 30 min
  11. 299How would you detect memorization or sensitive-data leakage from a model?Hard· 30 min
  12. 300How would you detect model drift in a generative-AI application?HardSystem design· 20 min
  13. 301How would you estimate the unit economics of an AI feature before launch?MediumProduct· 10 min
  14. 302How would you evaluate a multi-agent system?HardSystem design· 30 min
  15. 303How would you evaluate a RAG pipeline end to end?HardAI ML· 20 min
  16. 304How would you evaluate whether an enterprise AI pilot is ready for production?HardProduct· 20 min
  17. 305How would you explain hallucination risk to a regulated enterprise customer?MediumProduct· 10 min
  18. 306How would you handle citations and provenance in an AI research product?HardProduct· 20 min
  19. 307How would you implement semantic caching safely?HardSystem design· 20 min
  20. 308How would you improve the quality, latency, and cost of a AI developer platforms system without masking regressions?HardSystem design· 30 min
  21. 309How would you improve the quality, latency, and cost of a AI incident response system without masking regressions?HardSystem design· 30 min
  22. 310How would you improve the quality, latency, and cost of a AI watermarking system without masking regressions?HardSystem design· 30 min
  23. 311How would you improve the quality, latency, and cost of a content moderation system without masking regressions?HardSystem design· 30 min
  24. 312How would you improve the quality, latency, and cost of a continual learning system without masking regressions?HardSystem design· 30 min