Interview questions
Find a company or topic. Read the sources, then practice.
Loading the question library…
Find a company or topic. Read the sources, then practice.
Loading the question library…
Showing 251–268
The system must assess uncertain evidence, stop account misuse, and limit harm to legitimate users despite delayed confirmation of attacks.
General practice
activity-volume-and-timeliness · data-retention · delayed-outcomes · device-and-location-evidence
A model improvement loop with a fixed annotation budget, delayed human judgments, and uncertain gains.
General practice
active-learning · annotation · batch-processing · budget-control
Fast decisions and later fraud analysis must handle coordinated abuse, uncertain evidence, charge corrections, and errors against legitimate users.
General practice
event-time · false-positives · fraud-detection · idempotency
A click probability service must combine fresh features, durable prediction records, delayed labels, and calibrated models within a strict response deadline.
General practice
calibration · delayed-labels · feature-freshness · low-latency
Purchase probabilities with rare outcomes, delayed reports, incomplete attribution, and strict score latency.
General practice
attribution · class-imbalance · delayed-labels · feature-consistency
Agent execution across external tools: durable progress, uncertain tool results, permissions, and strict cost limits.
General practice
authorization · cost-control · durable-execution · failure-recovery
Code completions need relevant repository context, fast responses, and strict protection for private source code.
General practice
editor-context · edits-during-requests · repository-context · response-latency
Recorded audio to useful notes: long meetings, claims with evidence, access changes, and failures during costly model calls.
General practice
access-control · data-retention · note-accuracy · processing-latency
Caption accuracy, alignment with video time, and predictable completion times across languages, noise levels, and video lengths.
General practice
completion-time · multilingual-support · processing-workflow · repeated-requests-and-failures
Product relevance at low latency, with fresh behavior, new users, new products, and changes in product availability.
General practice
candidate-retrieval · cold-start · event-processing · experimentation
Email classification with costly false positives, delayed labels, hostile feedback, and attacks that change over time.
General practice
adversarial-feedback · concept-drift · data-privacy · email-security
Generate vectors at scale while keeping updates, deletions, and consumer copies correct during model changes.
General practice
changes-during-rebuilds · consumer-migration · embedding-generation · model-changes
Machine learning workflows with repeatable inputs, durable stage results, shared GPU capacity, and controlled model releases.
General practice
artifact-lineage · failure-recovery · gpu-scheduling · model-deployment
Model comparisons with reproducible experiments, fair judge scores, traceable human decisions, and bounded evaluation cost.
General practice
batch-processing · cost-control · experiment-reproducibility · human-review
A private model training service must share scarce GPUs, protect customer data, and recover long jobs without exposing incomplete artifacts.
General practice
artifact-storage · fine-tuning · gpu-scheduling · job-orchestration
GPU use under uneven load: first-token latency, token gaps, request memory, and fair client access.
General practice
capacity-planning · gpu-inference · latency-under-overload · mixed-length-request-scheduling
Concurrent experiments produce long metric histories and large artifacts. Reliable comparisons require exact records of code, data, and environments.
General practice
client-retries · data-lineage · large-artifacts · multi-tenancy
Model health checks with data drift, prediction drift, delayed labels, limits on quality claims, and alerts with useful evidence.
General practice
alerts · data-drift · delayed-labels · event-integrity