Public task directory

Technical interview practice tasks

Every entry below is a crawlable link to a unique public task page. This directory complements the interactive filters and keeps the complete catalog connected through standard pagination.

Page 185 of 386 · 18,491 tasks

Knowledge Check · Hard · 1 min

While reviewing Vue testing, a test mounts a costly third-party chart. Which action fixes it, and which check proves it worked

While reviewing Vue testing, a test mounts a costly third-party chart. Which action fixes it, and which check proves it worked?

View task

Knowledge Check · Hard · 1 min

While reviewing Vue testing, an assertion runs immediately after changing state. Which action fixes it, and which check proves it worked

While reviewing Vue testing, an assertion runs immediately after changing state. Which action fixes it, and which check proves it worked?

View task

Knowledge Check · Medium · 1 min

While reviewing Vue testing, a test calls a private component method. Which action best addresses the problem

While reviewing Vue testing, a test calls a private component method. Which action best addresses the problem?

View task

Knowledge Check · Medium · 1 min

While reviewing Vue testing, a row should emit its identifier on delete. Which action best addresses the problem

While reviewing Vue testing, a row should emit its identifier on delete. Which action best addresses the problem?

View task

Knowledge Check · Medium · 1 min

While reviewing Vue testing, a test mounts a costly third-party chart. Which action best addresses the problem

While reviewing Vue testing, a test mounts a costly third-party chart. Which action best addresses the problem?

View task

Knowledge Check · Medium · 1 min

While reviewing Vue testing, an assertion runs immediately after changing state. Which action best addresses the problem

While reviewing Vue testing, an assertion runs immediately after changing state. Which action best addresses the problem?

View task

ML Experiment Review · Medium · 40 min

A 14-day, three-arm homepage ranking experiment reached 2.81M users across two weekends. Candidate B leads aggregate NDCG, Candidate A has a smaller but cheaper lift, and the control remains strongest for new and sparse-history users. Audit

A 14-day, three-arm homepage ranking experiment reached 2.81M users across two weekends. Candidate B leads aggregate NDCG, Candidate A has a smaller but cheaper lift, and the control remains.

View task

ML Coding · Easy · 30 min

Split ids into consecutive batches of at most batch_size without dropping or reordering ids

Split ids into consecutive batches of at most batchsize without dropping or reordering ids.

View task

ML Coding · Medium · 30 min

Return true when any observed metric violates its max threshold; missing metrics do not trigger rollback

Return true when any observed metric violates its max threshold; missing metrics do not trigger rollback.

View task

ML Coding · Easy · 30 min

Given binary labels and probabilities, return fn, fp, tn, and tp at the supplied threshold

Given binary labels and probabilities, return fn, fp, tn, and tp at the supplied threshold.

View task

ML Coding · Easy · 30 min

Keep the first event for each event_id and return retained events in their original order

Keep the first event for each eventid and return retained events in their original order.

View task

ML Coding · Medium · 30 min

Return [prediction_id, label] pairs only when a label with the same id has arrived, preserving prediction order

Return [predictionid, label] pairs only when a label with the same id has arrived, preserving prediction order.

View task

ML Coding · Medium · 30 min

For each request return model_score when model_ok is true; otherwise return fallback_score. Preserve zero values

For each request return modelscore when modelok is true; otherwise return fallbackscore. Preserve zero values.

View task

ML Coding · Medium · 30 min

Return feature names whose age_seconds is greater than their max_age_seconds, sorted ascending

Return feature names whose ageseconds is greater than their maxageseconds, sorted ascending.

View task

ML Coding · Medium · 30 min

For each prediction, return the latest feature value for the same entity whose event_time is less than or equal to as_of. Preserve prediction order and return null when no feature is available

For each prediction, return the latest feature value for the same entity whose eventtime is less than or equal to asof. Preserve prediction order and return null when no feature is.

View task

ML Coding · Medium · 30 min

Merge ranked result lists, keep each id once at its highest score, then sort by descending score and ascending id

Merge ranked result lists, keep each id once at its highest score, then sort by descending score and ascending id.

View task

ML Coding · Medium · 30 min

Return a map from slice to [correct_count,total_count]. A row is correct when prediction equals label

Return a map from slice to [correctcount,totalcount]. A row is correct when prediction equals label.

View task

ML Coding · Easy · 30 min

Return the ids of the k highest-scoring candidates. Break equal scores by ascending id

Return the ids of the k highest-scoring candidates. Break equal scores by ascending id.

View task

ML Coding · Medium · 30 min

Documents are already ranked. Return the longest prefix of ids whose cumulative tokens do not exceed budget; skip no document inside the prefix

Documents are already ranked. Return the longest prefix of ids whose cumulative tokens do not exceed budget; skip no document inside the prefix.

View task

ML Experiment Review · Medium · 40 min

Fraud loss falls but false declines and manual-review load rise. Audit delayed outcomes, interference, asymmetric cost, and traffic ramp

Fraud loss falls but false declines and manual-review load rise. Audit delayed outcomes, interference, asymmetric cost, and traffic ramp.

View task

ML Experiment Review · Medium · 40 min

Resolution time falls, but hallucination review flags rise in one language. Audit judge reliability, user assignment, safety guardrails, cost, and scope

Resolution time falls, but hallucination review flags rise in one language. Audit judge reliability, user assignment, safety guardrails, cost, and scope.

View task

ML Experiment Review · Medium · 40 min

Open rate rises while unsubscribes increase for low-frequency users. Separate novelty from durable lift, audit exposure, and choose a lifecycle-specific launch

Open rate rises while unsubscribes increase for low-frequency users. Separate novelty from durable lift, audit exposure, and choose a lifecycle-specific launch.

View task

ML Experiment Review · Medium · 40 min

Conversion grows in aggregate but one region shows margin loss and allocation imbalance. Audit SRM, practical significance, compliance, and rollout

Conversion grows in aggregate but one region shows margin loss and allocation imbalance. Audit SRM, practical significance, compliance, and rollout.

View task

ML Experiment Review · Medium · 40 min

Long-term retention improves while short-term CTR falls. Audit metric hierarchy, novelty, creator concentration, and choose observation windows

Long-term retention improves while short-term CTR falls. Audit metric hierarchy, novelty, creator concentration, and choose observation windows.

View task

ML Experiment Review · Medium · 40 min

Answer acceptance improves, while citation coverage and latency regress for long documents. Audit evaluator agreement, slices, cost, and rollback

Answer acceptance improves, while citation coverage and latency regress for long documents. Audit evaluator agreement, slices, cost, and rollback.

View task

ML Experiment Review · Medium · 40 min

Search NDCG rises 3.1%, but zero-result queries and p95 latency regress. Audit assignment, power, segment interactions, cost, and define a gated rollout with rollback thresholds

Search NDCG rises 3.1%, but zero-result queries and p95 latency regress. Audit assignment, power, segment interactions, cost, and define a gated rollout with rollback thresholds.

View task

Model Incident · Medium · 40 min

Scores remain discriminative but become overconfident after a population shift, overloading manual review. Mitigate policy and rebuild monitoring

Scores remain discriminative but become overconfident after a population shift, overloading manual review. Mitigate policy and rebuild monitoring.

View task

Model Incident · Medium · 40 min

A catalog backfill changes item eligibility and new-user engagement falls while repeat-user metrics improve. Rank model, feature, and product causes

A catalog backfill changes item eligibility and new-user engagement falls while repeat-user metrics improve. Rank model, feature, and product causes.

View task

Model Incident · Medium · 40 min

Retrieval quality drops after the index and query encoder move to different embedding versions. Contain, verify, and prevent incompatible promotion

Retrieval quality drops after the index and query encoder move to different embedding versions. Contain, verify, and prevent incompatible promotion.

View task

Model Incident · Medium · 40 min

Conversion falls after a streaming lag causes stale features for one region while service health stays green. Rank evidence, mitigate safely, and add freshness gates

Conversion falls after a streaming lag causes stale features for one region while service health stays green. Rank evidence, mitigate safely, and add freshness gates.

View task

Model Incident · Medium · 40 min

GPU capacity loss routes traffic to a stale CPU fallback whose predictions are biased for one cohort. Restore service and quality together

GPU capacity loss routes traffic to a stale CPU fallback whose predictions are biased for one cohort. Restore service and quality together.

View task

Model Incident · Medium · 40 min

A prompt update improves completion rate but increases policy violations in one language. Stop harm, preserve evidence, and redesign release gates

A prompt update improves completion rate but increases policy violations in one language. Stop harm, preserve evidence, and redesign release gates.

View task

Model Incident · Medium · 40 min

False declines spike after a region-specific threshold configuration ships without model errors. Separate policy from model health and restore safely

False declines spike after a region-specific threshold configuration ships without model errors. Separate policy from model health and restore safely.

View task

ML Notebook Lab · Medium · 40 min

A default model gains 19 AUC points after charge-off fields are backfilled. Build a point-in-time split, isolate accounts, remove post-decision fields, and report future and thin-file slices

A default model gains 19 AUC points after charge-off fields are backfilled. Build a point-in-time split, isolate accounts, remove post-decision fields, and report future and thin-file.

View task

ML Notebook Lab · Medium · 40 min

A demand model performs well on shuffled rows but fails during promotions. Replace the random split with a chronological holdout, isolate stores, remove realized-promotion fields, and report cold-start slices

A demand model performs well on shuffled rows but fails during promotions. Replace the random split with a chronological holdout, isolate stores, remove realized-promotion fields, and report.

View task

ML Notebook Lab · Medium · 40 min

A fraud model treats unresolved transactions as negatives and uses investigation outcomes unavailable at authorization time. Build a matured-label temporal holdout and isolate cards

A fraud model treats unresolved transactions as negatives and uses investigation outcomes unavailable at authorization time. Build a matured-label temporal holdout and isolate cards.

View task

ML Notebook Lab · Medium · 40 min

An answer model is evaluated on prompts duplicated from training and uses post-review annotations as features. Build a time-aware prompt-family holdout and report unseen-domain quality

An answer model is evaluated on prompts duplicated from training and uses post-review annotations as features. Build a time-aware prompt-family holdout and report unseen-domain quality.

View task

ML Notebook Lab · Medium · 40 min

Repeated encounters put the same patient in train and validation while discharge-only fields leak the target. Build a patient-isolated temporal holdout and report hospital slices

Repeated encounters put the same patient in train and validation while discharge-only fields leak the target. Build a patient-isolated temporal holdout and report hospital slices.

View task

ML Notebook Lab · Medium · 40 min

A ranking model samples negatives from future impressions and repeats users across folds. Build a time-aware user-isolated holdout and report new-user quality

A ranking model samples negatives from future impressions and repeats users across folds. Build a time-aware user-isolated holdout and report new-user quality.

View task

ML Notebook Lab · Medium · 40 min

Augmented images from the same subject cross folds and a pathology-confirmation field arrives after capture. Build a subject-isolated temporal holdout and report device slices

Augmented images from the same subject cross folds and a pathology-confirmation field arrives after capture. Build a subject-isolated temporal holdout and report device slices.

View task

ML Project Deep Dive · Medium · 40 min

Explain a data or feature-quality failure, how you proved its model impact, the durable controls you built, and the outcome you personally delivered

Explain a data or feature-quality failure, how you proved its model impact, the durable controls you built, and the outcome you personally delivered.

View task

ML Project Deep Dive · Medium · 40 min

Explain an experiment where aggregate and slice evidence disagreed, the launch decision you owned, stakeholder trade-offs, and what happened next

Explain an experiment where aggregate and slice evidence disagreed, the launch decision you owned, stakeholder trade-offs, and what happened next.

View task

ML Project Deep Dive · Medium · 40 min

Explain why an apparently promising model did not ship, the evidence that changed your mind, sunk-cost handling, and how the learning transferred

Explain why an apparently promising model did not ship, the evidence that changed your mind, sunk-cost handling, and how the learning transferred.

View task

ML Project Deep Dive · Medium · 40 min

Present one model launch from problem contract through validation, rollout, impact, and a decision you personally owned. Include the baseline and a failed assumption

Present one model launch from problem contract through validation, rollout, impact, and a decision you personally owned. Include the baseline and a failed assumption.

View task

ML Project Deep Dive · Medium · 40 min

Explain model and retrieval choices, eval set construction, safety and cost constraints, production evidence, failures, and your owned decisions

Explain model and retrieval choices, eval set construction, safety and cost constraints, production evidence, failures, and your owned decisions.

View task

ML Project Deep Dive · Medium · 40 min

Explain a ranking or recommendation project, candidate and objective choices, evaluation, online effects, serving trade-offs, and your ownership

Explain a ranking or recommendation project, candidate and objective choices, evaluation, online effects, serving trade-offs, and your ownership.

View task

ML Project Deep Dive · Medium · 40 min

Explain a monitoring or reliability program, the failure mode it addressed, measurable change, rollout, and one alert or gate you removed or redesigned

Explain a monitoring or reliability program, the failure mode it addressed, measurable change, rollout, and one alert or gate you removed or redesigned.

View task

ML Project Deep Dive · Medium · 40 min

Explain a serving migration, capacity and latency budgets, compatibility plan, canary evidence, incident risk, and your direct contribution

Explain a serving migration, capacity and latency budgets, compatibility plan, canary evidence, incident risk, and your direct contribution.

View task