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
Filters · onDifficulty, topic, company
Clear all

2,498 questions

  1. 2473A pilot customer asks for a live dashboard, but the observed task is a weekly spreadsheet built from three exports. Define the next useful test.MediumProductSystem design
  2. 2474Two engineers serve ten pilots at five exports per minute. A proposal adds six microservices. What evidence would justify that cost?MediumSystem designTechnical
  3. 2475A report uses five calls at 0.02 units each and 0.03 compute units. Ten percent need two additional calls. Estimate average direct cost and one containment rule.MediumSystem designTechnical
  4. 2476A new version removes a field required by the previous application. The release fails. Why is code rollback alone insufficient?MediumSystem designTechnical
  5. 2477Thirty users are invited, eighteen start setup, twelve export successfully, and seven activated users return at the next relevant interval. All twelve activated users had a full follow-up window. Report the useful rates with limits.MediumDataSystem design
  6. 2478The worker process is healthy while oldest pending age rises. What should the only available engineer inspect and communicate?MediumBehavioralSystem design
  7. 2479Four engineer-days must cover an active access defect with a one-day containment-and-verification estimate, a failing paid-user workflow with a one-day diagnosis-and-fix estimate that remains uncertain, and a five-day prospect integration. Propose a feasible sequence and explain the opportunity cost.MediumProductSystem design
  8. 2480A save submits title Maple at version 7. Before it returns, the user types Birch and navigates to another record. Specify where the response may update state.MediumCodingTechnical
  9. 2481Two task creations overlap. The second succeeds, then the first fails validation. Why can restoring the original list corrupt the interface?MediumCodingTechnical
  10. 2482A server-rendered page serializes a full customer record into a client component and uses a project-only shared cache key. Identify two independent risks.MediumDataCoding
  11. 2483A job UI sees version 6, then version 4, then loses its event stream. Design recovery without restarting the job.MediumSystem designCoding
  12. 2484Design a deterministic test for an export whose acceptance response disappears after commit.MediumCodingTechnical
  13. 2485A page waits 80 ms for session, then serial independent calls of 220 and 350 ms, then renders for 70 ms. Estimate the benefit of parallel calls and state when it is invalid.MediumDataCoding
  14. 2486A model predicts at 09:00. A feature event occurred at 08:58 but arrived at 09:03. A backfill later rewrites its timestamp. How should the historical training row be built?MediumDataAI ML
  15. 2487A model's live scores change with the composition of each request batch. The estimator weights are identical to training. What transformation bug would you test?MediumAI MLTechnical
  16. 2488At one threshold TP=45, FP=55, FN=15, TN=885. False negatives cost 80 and false positives cost 3. Calculate precision, recall, and cost; state what else you need before deployment.MediumDataAI ML
  17. 2489A classifier has scores in [0, 1] and preserves its ranking after every score is squared. What can change in expected-cost decisions and how would you evaluate it?MediumDataAI ML
  18. 2490A colleague can reload weights but cannot reproduce the reported score. What experiment identity and first checks do you require?MediumAI MLTechnical
  19. 2491A momentum run resumes with correct weights but zero velocity. With w=2, v=3, next gradient=1, momentum=.9 and learning rate=.1, compare correct and weights-only next parameters. Use v_next = 0.9*v + g and w_next = w - 0.1*v_next, with no Nesterov, dampening, or weight decay.MediumDataAI ML
  20. 2492A vectorized masked loss matches output shape but changes training. Design a minimal reference fixture.MediumCodingAI ML
  21. 2493A 120 ms step spends 80 ms waiting for data and 40 ms in compute. A kernel halves compute time. What ideal speedup follows, and what trace check is necessary?MediumAI MLTechnical
  22. 2494Your optimization wins a benchmark only after excluding odd-length inputs. Those inputs are in the original contract. How should the result be reported?MediumAI MLTechnical
  23. 2495The best of 40 hyperparameter settings scores 88% on validation. A team wants to report 88% as final generalization. What should happen?MediumDataAI ML
  24. 2496A result estimates +0.3 points with an interval from -2 to +2.6, and useful improvement means at least +1. What can you conclude?MediumDataAI ML