ROLE
Data engineering
Compare expectations across career stages →
These are the same questions found in the library, grouped here by the capability each one mainly examines.
453 questions
Extended questions 401–450 of 453 on this page.
A · Professional problem solving and implementation28
- A new assistant improves satisfaction ratings, but task completion and repeat-contact rates remain unchanged. How would you determine whether this is a real improvement?
- Model responses may echo customer identifiers, while production debugging relies on request and response logs. How would you design collection, redaction, access, and retention?
- Users see response times double, but model inference time has barely changed. How would you break down the request-processing stages, locate the bottleneck, and limit impact?
- The team tunes prompts and model settings against the same fixed evaluation set every week. Scores rise, but new-user feedback does not improve. How would you restore evaluation trustworthiness?
- A model passes evaluations in a major market but behaves inconsistently in a new region because language, terminology, and service rules differ. How would you set rollout scope and follow-up gates?
- To reduce inference cost, the team caches answers by prompt text. User permissions, knowledge versions, and regional rules may differ. How would you define cache keys and invalidation?
- The evaluation set has grown, but many items are slight rewrites of one template, making model scores look stable. How would you audit sample quality?
- A shared cache is read-heavy and the team plans to replace a mutex with a reader-writer lock. What evidence would you gather, and how would you judge whether the added complexity is worthwhile?
- A migration across millions of records cannot fit in one transaction. The business cannot expose a partially completed state. How would you design batched writes, visibility, and recovery?
- A service accumulates monetary values with floating-point numbers, and reconciliation drift grows over time. How would you identify the source and redesign calculations while migrating historical data?
- A dashboard flags anomalies but shows no magnitude, baseline, or related events. Users export data to interpret it themselves. How would you improve diagnosis?
- A template changes only formatting and examples, but output length and tool-call frequency shift. The team wants to skip full regression. How would you tier the risk?
- A new safety policy reduces false refusals but causes sensitive disclosures in a few edge cases. Product wants to keep the satisfaction gain. How would you evaluate?
- A hosted model upgrades without notice. Standard health checks pass, but classification boundaries and output length shift. How would you build safeguards?
- After a safety policy launch, human handoffs double while high-risk errors fall. Operations capacity is limited. How would you assess the result and tune thresholds?
- A model labels some accounts high-risk and automatically denies service, affecting customer rights. How would you set validation, appeals, and human review?
- The business changes “churn” from 60 days without login to 30 days without payment, while historical training and models use the old definition. How would you manage versions and comparisons?
- Given many timestamped records, return the top K by score and preserve earlier items on ties. How would you implement it and analyze edge cases?
- Several sources are individually time-sorted, but total data exceeds memory. Explain how to merge output and handle a source that slows or disconnects temporarily.
- CPU averages 35%, but the request queue keeps growing and threads wait on external I/O. How would you identify the real capacity bottleneck?
- Database storage grows while cleanup scans cannot reclaim space; reporting transactions run for hours. How would you diagnose and govern this?
- Instance health checks verify only process liveness, while user requests fail on a critical dependency timeout. Adding every dependency to probes may remove all instances. How would you design this?
- Multiple regions accept the same request independently. Idempotency records have a TTL and machine clocks may differ by seconds. How would you prevent duplicate writes at expiry boundaries?
- A live index needs a new shard key as data grows quickly and writes cannot stop. How would you plan old and new indexes and decide when to cut over?
- A multi-tenant query platform shares a database pool. One tenant submits many complex reports and increases latency for others. How would you isolate and meter usage?
- A user starts a large export and disconnects; access is revoked while the job runs. The job needs resumability without leaking data after revocation. How would you design it?
- A live dashboard uses event-time windows, while finance groups by settlement time and excludes unsettled transactions. A leader believes one report must be wrong. How would you explain and govern this?
- Every read request is logged, driving cost up, while compliance requires evidence for sensitive actions. How would you tier collection and retention?
No questions in this section match your target. ·
B · Business understanding and domain modeling11
- Revenue was recognized for an order last month, and a partial refund occurs this month. Finance and growth disagree on order count, net revenue, and refund-month treatment. How would you establish consistent data semantics?
- A service rebuilds state from an event log, but the live snapshot differs and responders cannot tell which to repair. How would you define authority and the repair process?
- After a feature definition is fixed, the modeling team wants a historical backfill. Training reproducibility requires the original version, while business reports want corrected values. How would you design both views?
- Payment is authorized but not captured, and goods are not fulfilled. The business report counts an order, while finance records no revenue yet. How would you define and reconcile order, payment, revenue, and cash metrics across domains?
- Risk teams need to reproduce decision inputs from three months ago, but the online feature store keeps only the latest state to save cost. How would you balance reproducibility and storage cost?
- A missing risk feature is automatically set to zero. Latency and error rates are stable, but denial rates changed significantly. How would you judge whether the default is safe?
- Operations counts revenue when an order is placed, while finance recognizes it after fulfillment and refunds. A leader wants one number. How would you explain the two measurement methods and design shared measures for decisions?
- Several teams manually corrected an aggregate table, and later no one could explain why the number changed. How would you record sources, transformations, versions, owners, and recomputation evidence?
- A product line splits into two business units. Old orders belong to the previous organization, while future orders use the new structure. How would you represent organization versions and historical views?
- A proposal repeatedly refers to “core users” without identification rules, observation windows, or exclusions. What would you clarify so the team can study the same population?
- Marketing proposes behavior-based discounts that may raise revenue, but customers can compare offers. How would you define pricing rules, disclosure, and risk monitoring?
No questions in this section match your target. ·
C · Value judgment and direction setting3
- During a randomized experiment, the platform changes pricing policy and affects treatment and control differently. How would you decide which conclusions the experiment still supports?
- Brand advertising expands reach, but short-term direct conversions do not change and the sales cycle may last months. How would you decide whether to wait, change creative, or stop?
- A recommendation algorithm changes content ranking and creator supply, so treatment behavior changes what the control group can see. How would you assess the experiment design?
No questions in this section match your target. ·
D · Delivery and ongoing operations1
No questions in this section match your target. ·
E · Collaboration, communication, and influence1
No questions in this section match your target. ·
H · Professional responsibility and work style6
- At a quarterly review, you predicted a change would lift conversion. A later segmented experiment shows no overall lift and harm to one segment. How would you correct the conclusion?
- Tell me about a time you proactively corrected something you had said or a metric definition you had used. How did you find it, and how did you tell the people affected?
- You find a colleague omitted one class of refunds from business results already sent to leadership. They suggest quietly replacing the file. How would you ensure the conclusion and decision record are corrected?
- Your quarterly report says a target was met. You later discover that the data pipeline changed the definition of “active customer” midway through the period; recalculating with the old definition puts the result below target. How would you correct the report, inform decision-makers, and repair the process?
- Describe a project or module you owned that fell short of its commitment. How did dependencies and unvalidated assumptions affect the result, what was your responsibility, and how did you recover and improve your validation for next time?
- An experiment report shows target conversion, but a review finds that the query includes only users who completed the flow. A leader is about to request scale-up. How would you respond?
No questions in this section match your target. ·