← Etsy Interview Insights

Etsy·Data Scientist·Onsite - System Design / Architecture·Senior

Senior
Jul 2026

Summary

Etsy data scientist interview that was basically one very long system design question about search autocomplete ranking. The scope was massive and I kept second-guessing how deep to go on each part.

Questions Asked (7)

Q1

Given query and click log data with fields like user ID, timestamp, locale, device, typed prefix, suggested term, position, click indicator, dwell time, downstream query, and eventual success, design an ML system to re-rank autocomplete candidate suggestions for each prefix.

System DesignTechnical Trade-offsProduct Analytics & Metrics
Author's notes

This was a lot.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business objective and defining success metrics, then frame the problem as a learning-to-rank task using the click log data. Walk through the ML pipeline from data preparation and feature engineering to model selection, training, and online evaluation, emphasizing how you would handle biases and cold-start issues.

Pro tip: Highlight that click data is biased by position and presentation, so you need to correct for position bias (e.g., using inverse propensity scoring or a position-based model) to avoid feedback loops. Also, mention that offline metrics like NDCG may not correlate with online business metrics, so plan for rigorous A/B testing.

1. Clarify Objective and Metrics

Define the goal (e.g., increase successful completions, reduce abandonment) and choose metrics like MRR, NDCG, or business KPIs (e.g., conversion rate, dwell time).

2. Data Preparation and Labeling

Aggregate click logs per query-prefix and candidate, define relevance labels (e.g., click, dwell time, downstream success), and handle biases like position and presentation.

3. Feature Engineering

Create features from user (locale, device, history), query (prefix, popularity), candidate (term frequency, semantic similarity), and context (time, session).

4. Model Selection and Training

Choose a learning-to-rank model (e.g., LambdaMART, neural ranker) and train with unbiased labels, using techniques like propensity weighting or counterfactual learning.

5. Evaluation and Deployment

Evaluate offline with ranking metrics and online via A/B tests, then deploy with monitoring for drift and feedback loops.

Key Points to Mention

  • Position bias correction (e.g., inverse propensity scoring, position-based model)
  • Choice of ranking model (pointwise, pairwise, listwise) and rationale
  • Feature engineering from user, query, candidate, and context
  • Handling cold-start and sparsity (e.g., fallback to popularity or content-based)
  • Offline vs. online evaluation metrics and A/B testing
  • Feedback loops and long-term effects (e.g., exploration vs. exploitation)

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

What labels best capture long-term user success in an autocomplete system, and how do you construct train/validation splits that avoid temporal leakage?

Data ModelingA/B Testing & Experimentation
Author's notes

Said eventual_success and session-level success over raw click, which the interviewer seemed fine with.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining long-term user success in autocomplete as sustained engagement and satisfaction, not just immediate clicks. Then explain how to construct temporal splits that respect the time order of interactions, using time-based cutoffs and ensuring no future data leaks into training. Emphasize the importance of aligning labels with business goals and validating with time-aware metrics.

Pro tip: Use a rolling-origin evaluation scheme to simulate real-world deployment, and consider delayed feedback windows to capture long-term success. This shows you understand production constraints and avoid overfitting to short-term signals.

1. Define long-term success labels

Identify labels that reflect sustained user value, such as repeat usage, session completion, or long-term retention, rather than immediate clicks. Consider business-specific metrics like purchase conversion or search success.

2. Choose temporal split strategy

Split data by time (e.g., train on older data, validate on newer) to mimic real deployment. Use a cutoff date and ensure no future data is used in training.

3. Handle delayed feedback

Account for labels that mature over time (e.g., a user returning after a week). Use a feedback window and only include examples where the outcome is fully observed.

4. Validate with time-aware metrics

Evaluate using metrics that reflect long-term success, such as retention rate or cumulative gain, and compare against baselines. Use rolling windows to assess stability.

5. Iterate and monitor

Continuously update splits as new data arrives and monitor for drift. Revisit label definitions to ensure they still capture long-term success.

Key Points to Mention

  • Temporal leakage: using future data in training leads to overly optimistic performance.
  • Long-term success labels: retention, repeat usage, session success, purchase conversion.
  • Time-based splitting: train on past, validate on future, with a gap for delayed feedback.
  • Rolling-origin evaluation: multiple train/validate splits over time to simulate deployment.
  • Delayed feedback: labels may not be immediately available; use a maturity window.
  • Business alignment: labels should tie to Etsy's goals like repeat purchases or seller engagement.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

How would you correct for position and selection bias in the click logs, for example through counterfactual logging, inverse propensity weighting, or randomized interleaving?

A/B Testing & ExperimentationTechnical Trade-offs
Author's notes

Knew the concepts but my answer got circular.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining position and selection bias in click logs and why they matter for ranking and recommendation systems. Then compare counterfactual logging, inverse propensity weighting (IPW), and randomized interleaving in terms of assumptions, data requirements, and trade-offs. Finally, recommend a practical approach for Etsy, such as combining IPW with randomized interleaving for unbiased evaluation, and discuss implementation challenges.

Pro tip: Emphasize that no single method is perfect; the best approach depends on the stage of experimentation and production constraints. Show awareness that IPW can have high variance and that randomized interleaving requires careful design to avoid user disruption.

1. Define the problem

Explain what position and selection bias are in click logs: users click on items because of their position or because they were selected by a biased policy, not necessarily because they are more relevant.

2. Introduce correction methods

Briefly describe counterfactual logging (logging randomized results to create an unbiased dataset), inverse propensity weighting (weighting clicks by the inverse probability of being shown), and randomized interleaving (mixing results from two policies to compare them fairly).

3. Compare trade-offs

Discuss assumptions, data requirements, and practical challenges: IPW needs accurate propensity scores and can be high variance; counterfactual logging requires infrastructure to log randomized results; interleaving requires user-facing randomization and careful metric design.

4. Recommend an approach

Suggest a combined strategy: use randomized interleaving for online evaluation of new models, and IPW or counterfactual logging for offline evaluation and training unbiased models. Tailor to Etsy's scale and constraints.

5. Address implementation

Mention practical considerations: logging infrastructure, propensity model accuracy, variance reduction techniques (e.g., clipping), and monitoring for bias drift over time.

Key Points to Mention

  • Position bias: users click on top-ranked items regardless of relevance.
  • Selection bias: the logging policy only shows items it favors, so clicks are not representative.
  • Inverse propensity weighting (IPW): reweight clicks by the inverse probability of being shown to correct for bias.
  • Counterfactual logging: log randomized results to create an unbiased dataset for offline evaluation.
  • Randomized interleaving: mix results from two policies in a single session to compare them without bias.
  • Trade-offs: IPW can have high variance; interleaving requires user-facing randomization and careful metric design; counterfactual logging needs infrastructure.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

What feature sets would you use, covering contextual signals, lexical features, popularity time series, and semantic embeddings, and how do you handle multilingual text and Unicode normalization?

Data ModelingSystem Design
Author's notes

Talked through prefix edit distance, session context, and time-decayed popularity counts without much trouble.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around the four feature categories, explaining how each contributes to search relevance and ranking at Etsy. Then discuss multilingual handling and Unicode normalization as critical preprocessing steps that ensure feature consistency across languages. Emphasize trade-offs and practical implementation considerations.

Pro tip: Tie features to business metrics like conversion or engagement, and mention how you'd validate multilingual handling with A/B tests or offline evaluation. Show awareness of Etsy's marketplace dynamics, such as long-tail queries and diverse seller content.

1. Contextual Signals

Describe features like user location, device, time of day, and session history that personalize results. Explain how they capture intent and context for ranking.

2. Lexical Features

Cover text-based features such as TF-IDF, BM25, n-grams, and exact match indicators. Highlight their role in matching query terms to item titles and descriptions.

3. Popularity Time Series

Discuss temporal features like recent views, favorites, sales velocity, and seasonality. Explain how they capture trends and item freshness.

4. Semantic Embeddings

Introduce dense vector representations (e.g., from BERT or sentence transformers) to capture semantic similarity beyond exact matches. Mention how they handle synonyms and paraphrases.

5. Multilingual & Unicode Handling

Explain strategies for multilingual text: language detection, translation, or multilingual embeddings. Detail Unicode normalization (NFC/NFD), case folding, and handling diacritics to ensure consistent tokenization.

Key Points to Mention

  • Feature engineering pipeline: how features are computed, stored, and updated in real-time or batch.
  • Multilingual embeddings (e.g., LASER, mBERT) and cross-lingual transfer for low-resource languages.
  • Unicode normalization forms (NFC, NFD, NFKC, NFKD) and their impact on tokenization and matching.
  • Handling of emojis, special characters, and mixed-script text common in user-generated content.
  • Evaluation metrics for search relevance (NDCG, MRR) and how to measure multilingual performance.
  • Scalability considerations: feature serving latency, storage, and computational cost.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q5

What model architecture would you choose given latency and memory constraints at serving time, and what's your fallback strategy for cold-start terms or new users?

System DesignTechnical Trade-offs
Author's notes

Went with a two-stage setup: a fast retrieval layer then a lightweight ranker, something like a small gradient boosted model rather than a neural net for latency reasons.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the serving constraints (latency SLA, memory budget, throughput) and the business context (Etsy's marketplace with long-tail queries and new users). Then propose a two-tower retrieval model with a lightweight ranking model, explaining how it balances quality and efficiency, and outline a fallback strategy that blends popularity, content-based, and session-based signals for cold-start cases.

Pro tip: Quantify trade-offs with concrete numbers (e.g., 'two-tower with 64-d embeddings fits in X GB and serves in Y ms') and mention that you'd monitor online metrics like CTR and conversion to validate the fallback's impact.

1. Clarify constraints and context

Ask about latency SLA, memory limits, QPS, and the nature of cold-start (new terms vs. new users). Confirm the business goal (e.g., relevance, diversity, revenue).

2. Propose a primary architecture

Recommend a two-tower retrieval model (user and item towers) with approximate nearest neighbor search, followed by a lightweight ranking model (e.g., GBDT or small MLP). Explain how it meets latency and memory constraints.

3. Address cold-start fallback

Describe a fallback pipeline: for new terms, use content-based embeddings (e.g., from text) or query expansion; for new users, use popularity, session-based, or demographic-based recommendations. Emphasize graceful degradation.

4. Discuss trade-offs and alternatives

Compare with other architectures (e.g., single-stage ranker, matrix factorization) and explain why your choice is optimal given constraints. Mention potential hybrid approaches.

5. Outline evaluation and iteration

Explain how you'd measure success offline (recall@k, NDCG) and online (A/B tests on CTR, conversion). Describe how you'd iterate on the fallback strategy based on performance.

Key Points to Mention

  • Two-tower retrieval + lightweight ranking for efficient serving
  • Approximate nearest neighbor (ANN) for fast retrieval
  • Cold-start fallback: content-based, popularity, session-based, or hybrid
  • Latency and memory trade-offs: embedding dimensions, model size, quantization
  • Online metrics (CTR, conversion) and offline metrics (recall, NDCG) for evaluation
  • Graceful degradation and monitoring of fallback usage

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q6

How would you limit feedback loops, concept drift, and surfacing of unsafe or low-quality suggestions over time?

Product StrategyTechnical Trade-offs
Author's notes

Talked about periodic retraining, monitoring distribution shift on query logs, and a blocklist plus quality classifier for unsafe terms.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around a continuous monitoring and feedback system that balances model performance with safety and quality. Emphasize proactive detection of drift and unsafe outputs, and describe how you would incorporate human-in-the-loop and automated safeguards to limit feedback loops. Tailor your response to Etsy's context by highlighting the importance of seller and buyer trust, and the need for scalable solutions in a dynamic marketplace.

Pro tip: Demonstrate awareness of the trade-offs between rapid iteration and long-term safety by proposing a tiered alerting system that distinguishes between minor drift and critical safety violations, ensuring that high-risk issues are escalated immediately while low-risk ones are monitored for trends.

1. Define Metrics and Guardrails

Establish clear metrics for model performance, concept drift, and safety/quality thresholds. Define what constitutes an unsafe or low-quality suggestion and set up automated alerts when these thresholds are breached.

2. Implement Continuous Monitoring

Deploy real-time monitoring for data drift, prediction drift, and feedback loops. Use statistical tests and dashboards to track changes in input distributions and model outputs over time.

3. Incorporate Human-in-the-Loop and Feedback Mechanisms

Design systems for human review of flagged outputs and user feedback (e.g., reporting unsafe suggestions). Use this feedback to retrain and improve models, while avoiding feedback loops by diversifying data sources.

4. Mitigate Drift and Feedback Loops

Apply techniques like regular retraining, ensemble methods, and causal inference to reduce concept drift. Break feedback loops by introducing exploration (e.g., epsilon-greedy) and using unbiased data collection.

5. Iterate and Govern

Establish a governance process for model updates, including A/B testing for safety and quality. Continuously refine safeguards based on incidents and evolving business needs.

Key Points to Mention

  • Concept drift detection using statistical process control and windowed comparisons
  • Feedback loop mitigation via exploration/exploitation trade-offs and unbiased sampling
  • Safety and quality filters: rule-based, ML-based, and human moderation
  • Human-in-the-loop for edge cases and continuous learning
  • Monitoring dashboards and alerting for proactive issue resolution
  • Governance and cross-functional collaboration (e.g., with policy, legal, and product teams)

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q7

Walk through your offline and online evaluation plan for this system, including what rollback criteria you would set.

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

Offline: NDCG on held-out sessions, coverage on tail queries.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by framing the evaluation plan around the system's objectives and key metrics, then detail offline validation methods and online A/B testing design. Conclude with specific rollback criteria tied to guardrail metrics and business impact.

Pro tip: Emphasize that rollback criteria should be pre-registered and include both statistical significance and practical significance thresholds to avoid false alarms and ensure meaningful impact.

1. Define Objectives and Metrics

Clearly state the system's goal and identify primary success metrics (e.g., CTR, conversion) and guardrail metrics (e.g., latency, error rate, user satisfaction).

2. Offline Evaluation

Describe how you would validate the system offline using historical data, cross-validation, and simulated environments, ensuring no leakage and representative sampling.

3. Online Evaluation Design

Outline the A/B test setup: randomization unit, sample size calculation, test duration, and how you'll monitor metrics in real-time.

4. Rollback Criteria

Specify quantitative thresholds for rollback, such as a statistically significant drop in a guardrail metric or a negative impact on a key business metric beyond a predefined tolerance.

5. Monitoring and Iteration

Explain how you'll monitor the experiment, analyze results, and decide whether to roll out, iterate, or roll back, including post-mortem analysis if rollback occurs.

Key Points to Mention

  • Use of guardrail metrics to ensure system health and user experience
  • Pre-registration of hypotheses and rollback criteria to avoid p-hacking
  • Sample size and power analysis to detect meaningful effects
  • Consideration of novelty effects and long-term impact
  • Automated alerting for real-time monitoring of rollback triggers
  • Business impact assessment and stakeholder communication plan

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.