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 0Medium 24Hard 7
Filters · onDifficulty, topic, company

Difficulty

Clear all

65 questions

  1. 25Prove logistic loss has a global minimumData· Amazon
  2. 26Prove MSE is non-convex for logistic regressionData· Microsoft
  3. 27Technical deep-dive on an issue (Dive Deep)HardData· Amazon
  4. 28The project you're most proud ofData· Amazon
  5. 29Urgent decision with incomplete information (Bias for Action)MediumData· Amazon
  6. 30Why is a GPU faster than a CPU? Why is NumPy faster than a list?Data· Microsoft
  7. 31Your ML approach didn't workMediumData· Microsoft
  8. 32Classify products into categoriesData· Amazon
  9. 33Closed-form linear regression vs gradient descentData· Amazon
  10. 34Code conv2d from scratchData· Microsoft
  11. 35Cohort funnel in pandasMediumData· PDPinterest DS take-homes
  12. 36Derive the logistic loss on a whiteboardHardData· Pinterest
  13. 37Fine-tune a pre-trained LLM for a product featureData· Microsoft
  14. 38How did you debug your ML project?HardData· Amazon
  15. 39KL divergence vs cross-entropyData· Amazon
  16. 40Merge intervals plus ML fundamentals probeMediumData· Pinterest
  17. 41Parse data and compute statisticsMediumData· Scale AI
  18. 42Prompt vs fine-tune vs RAG for a search-and-answer systemMediumData
  19. 43Prove 1-NN error is at most twice the Bayes errorData· Microsoft
  20. 44Push notifications before conversionMediumData· Uber
  21. 45Sequential tests and simultaneous experimentsData· Netflix
  22. 46Stadium with 3+ consecutive high-attendance days (SQL)MediumData· Netflix
  23. 47Statistical methods you've actually usedData· Amazon
  24. 48Supervised vs unsupervised vs reinforcement learningData· Amazon