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. 2425Why do you want to work at Rivian?MediumBehavioral
  2. 2426Why do you want to work at Hinge?MediumBehavioral
  3. 2427Review System76's public Backend Engineer product-query API assignment. Explain a plan for data modeling, category filtering, verification, and a concise assumptions note.MediumCodingSystem design· SSystem76
  4. 2428Account A starts at 100 units before this batch. Each accepted event adds its signed delta; these are not full balance snapshots. Durable event identity is (source, event_id), and the first valid accepted payload is immutable. An exact retry has no new effect. A later different payload with the same identity is quarantined and does not replace the accepted event. Corrections must be new authorized events with their own identity and a reference to the original event. All rows below target A and the correction is authorized. Process them in the stated delivery order.MediumDataTechnical
  5. 2429Use Monzo's historical public web-engineer portal task as a practice reference. Explain how you would handle token expiry, app updates, and paginated user lists without stale cross-app state.MediumCodingTechnical· Monzo
  6. 2430A search component keeps the last successful results visible while a newer search is pending. Each request has an increasing request ID. Its success handler checks whether the ID is still active, but its failure and cleanup handlers do not:MediumCodingTechnical
  7. 2431A member of tenant red uploads an untrusted document. An isolated parser extracts text and proposes a title. A red reviewer previews the result; only a red publisher may release a reviewed version. The parser cannot access the network and has a per-job CPU, memory and output-size limit. Storage keeps immutable upload and extraction versions. The preview renders extracted content. Name the trust boundaries, two threats that remain after parser isolation, and negative and positive tests. A parser result must never grant publication authority.MediumTechnical
  8. 2432A 24 GiB GPU holds13 GiB of weights and reserves3 GiB for runtime. Standard non-MLA attention has 32 layers,8 KV heads,128 head dimension and 2-byte cache values. Every request may hold8,192 input-plus-output tokens. Calculate the simplified concurrency bound and explain why a test at twelve concurrent requests can fail.MediumAI MLSystem design
  9. 2433The same mixed chat/document trace produces 900 output tokens/s with chat p95 gaps 35 ms. A larger batch produces 1,080 tokens/s but chat gaps 70 ms. The chat limit is 50 ms. Decide whether to accept and propose a controlled next test.MediumAI MLSystem design
  10. 2434Eight GPUs are available on four hosts, two per host. The model requires two GPUs per replica. A host-local replica measures40 requests/minute; an eight-GPU group measures110. Demand is 100/minute after one host loss. Compare the two layouts under the assumption that a parallel replica needs every worker.MediumAI MLSystem design
  11. 2435In matched one-hour measurement runs, configuration A costs 12 units/hour and serves 18,000 completed responses, of which 3,000 are late. Configuration B costs 10 units/hour and serves 12,000 timely responses. No other failures occur. Which is cheaper per thousand timely successes, and what does that comparison omit?MediumAI MLSystem design
  12. 2436A shared pool has 12,000 free reserved tokens; tenant T has 4,000 under its active limit. A request needs 6,000 and expires in one second; no release is expected for three seconds. What should admission do, and how should a retry reuse identity?MediumAI MLSystem design
  13. 2437Two replicas each serve30 requests/minute. Arrivals rise to 90. A third replica takes three minutes to become ready; nothing expires or is rejected. Calculate backlog at readiness and whether it drains. What if a fourth is ready then?MediumAI MLSystem design
  14. 2438A consumer calls an external credit API for operation e42. The API applies10 units, but its response is lost before the consumer commits its broker offset. Does broker exactly-once processing prevent a second credit? Design the retry boundary.MediumDataTechnical
  15. 2439Window[12:00,12:10) UTC has values2 at 12:00 and 5 at 12:09. A value 7 at 12:10 belongs to the next interval. An accepted late value 4 occurred12:03. What is the revised first-window sum, how is it published, and what does a watermark not prove?MediumDataTechnical
  16. 2440A three-day replay expects counts100,122,79 after a known parser repair. Candidate output is 101,121,79. The total matches301. Should it replace production, and what must make a second run safe?MediumDataTechnical
  17. 2441A stopped CDC connector retains2 GiB of source log per hour. The source has 8 GiB free and no other disk use in this simplified model. Explain the deadline and the risk of discarding retained logs to free space.MediumDataTechnical
  18. 2442A daily invoice table should contain one row per invoice and required customer references. It has 1,000 rows, a recent timestamp,20 duplicate keys and one missing source partition. Define publication behavior and evidence.MediumDataTechnical
  19. 2443A Deployment desires3 replicas;3 Pods exist, only 1 is ready, and clients see failures through its Service. What evidence do you collect before changing replica count?MediumSystem designTechnical
  20. 2444A node has 8 allocatable CPUs and existing ordinary-container requests total 7.2. A pending Pod requests 0.7 for its app and 0.3 for a sidecar. Live CPU is 20%. Assume no init or Pod-level resource effects. Why can it remain Pending?MediumSystem designTechnical
  21. 2445A healthy API process loses its shared database. Cached catalogue reads still work, but writes fail. The current liveness probe checks the database and repeatedly restarts every Pod. What policy would you propose?MediumSystem designTechnical
  22. 2446A destination Pod is selected by two ingress policies: one allows only app=api and another allows all Pods in the namespace. A test Pod in that namespace connects. The network provider enforces policy. Why?MediumSystem designTechnical
  23. 2447A Deployment has 4 replicas, maxUnavailable=0 and maxSurge=1. Nodes have no spare requested capacity for the surge Pod. Old replicas meet demand only while all four serve. Does this prove a zero-downtime rollout?MediumSystem designTechnical
  24. 2448Five comparable developers try to create and deploy a service. Four complete unassisted in20,25,30,35 minutes; one finishes in60 minutes with assistance. What platform outcome do you report and test next?MediumSystem designTechnical