← Amazon Interview Insights

Amazon·Research Scientist·Onsite - System Design / Architecture·Senior

SeniorPrefer not to say
Jun 2026

Summary

Amazon RS loop, ML system design focused on search ranking and experimentation. The round goes deep on a single axis rather than broad coverage, which I wasn't fully expecting. Generative variants with brand-fidelity checks made it more interesting than a standard ranking question.

Questions Asked (5)

Q1

Walk through how you would design a search ranking system end to end, from retrieval through re-ranking.

System DesignTechnical Trade-offs
Author's notes

The key thing nobody tells you before these rounds: they don't want breadth, they want you to pick a layer and go five levels deep.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the problem scope and requirements, then walk through the system in stages: retrieval, ranking, and re-ranking. For each stage, explain the design choices, trade-offs, and how you would evaluate and iterate on the system.

Pro tip: Emphasize the importance of offline and online evaluation metrics, and how you would use A/B testing to validate improvements. Also, discuss how you would handle cold-start and scalability challenges, showing awareness of production constraints.

1. Clarify Requirements and Scope

Ask questions to understand the scale, latency requirements, data available, and business objectives. Define success metrics such as NDCG, MRR, or business KPIs like conversion rate.

2. Design Retrieval Stage

Explain how to efficiently fetch a candidate set from a large corpus using techniques like inverted indexes, dense retrieval with embeddings, or hybrid approaches. Discuss trade-offs between recall and latency.

3. Design Ranking Stage

Describe how to score and order candidates using machine learning models (e.g., LambdaMART, neural rankers). Cover feature engineering, model training, and handling of position bias.

4. Design Re-ranking Stage

Explain how to refine the top results using more expensive models or business rules, considering diversity, freshness, and personalization. Discuss trade-offs between relevance and other objectives.

5. Evaluation and Iteration

Outline offline evaluation with historical data and online evaluation via A/B testing. Discuss how to monitor system health, detect degradation, and iterate on models and features.

Key Points to Mention

  • Two-stage or multi-stage ranking architecture (retrieval, ranking, re-ranking)
  • Trade-offs between recall and precision, latency and quality
  • Use of embeddings and approximate nearest neighbor search for retrieval
  • Learning to rank algorithms and feature engineering
  • Handling position bias and counterfactual evaluation
  • Online metrics (CTR, conversion) and offline metrics (NDCG, MRR)

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

Q2

How would you evaluate generation quality for image or text outputs?

Product Analytics & MetricsA/B Testing & Experimentation
Author's notes

Broader than it sounds.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by acknowledging that generation quality is multi-dimensional and depends on the task, then outline a layered evaluation strategy that combines automated metrics, human judgment, and online experimentation. Emphasize how you would choose metrics aligned with business objectives and validate them through A/B tests to ensure real-world impact.

Pro tip: Don't just list metrics—explain how you'd validate them against human judgments and business KPIs, and mention the importance of monitoring for distribution shifts and bias in production.

1. Define quality dimensions and objectives

Identify what 'quality' means for the specific task (e.g., fidelity, coherence, relevance, diversity, safety) and align with business goals. Consider both intrinsic (output-only) and extrinsic (downstream task) evaluations.

2. Select automated metrics and models

Choose appropriate automatic metrics (e.g., BLEU, ROUGE, BERTScore, FID, IS for images) and learned evaluators (e.g., trained classifiers, LLM-as-a-judge) that correlate with human perception. Be transparent about their limitations.

3. Incorporate human evaluation

Design human studies with clear rubrics (e.g., Likert scales, pairwise comparisons) and ensure annotator quality via training and agreement checks. Use human judgments as ground truth to calibrate automated metrics.

4. Run offline and online experiments

Conduct offline evaluations on held-out sets, then deploy A/B tests to measure impact on user engagement and business metrics. Use guardrail metrics to catch regressions in safety or fairness.

5. Monitor and iterate in production

Continuously track quality metrics, detect drift, and collect user feedback to refine models and evaluation pipelines. Close the loop by feeding insights back into model development.

Key Points to Mention

  • Task-specific metrics: e.g., BLEU/ROUGE for text, FID/IS for images, but note their limitations.
  • Human evaluation: importance of well-designed studies, annotator agreement, and pairwise comparisons.
  • A/B testing: measuring impact on user engagement and business KPIs, with guardrail metrics.
  • LLM-as-a-judge: using large language models for scalable evaluation, but validating against human judgments.
  • Bias and fairness: evaluating for representational harm and ensuring safety.
  • Production monitoring: detecting distribution shifts and incorporating user feedback loops.

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

Q3

If an image is generated for a brand ad, how do you ensure the brand name appears correctly in the output?

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

This one is sneakier than it looks.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Frame the problem as a system design challenge: ensuring brand name correctness in generated images requires a multi-layered approach combining model conditioning, post-generation verification, and fallback mechanisms. Discuss trade-offs between generation quality, latency, and accuracy, and propose metrics to evaluate success.

Pro tip: Emphasize that perfect text generation is still an open problem; a robust system should include a verification step (e.g., OCR) and a fallback to overlay the brand name programmatically if the generated text is incorrect. This shows practical maturity and risk mitigation.

1. Clarify Requirements and Constraints

Ask about the use case: Is the brand name required to be perfectly legible? What is the acceptable latency? Are there multiple brand names or languages? This scopes the problem and informs trade-offs.

2. Leverage Conditional Generation

Use models that accept text conditioning (e.g., diffusion models with text encoders) and fine-tune on brand-specific data to improve fidelity. Consider incorporating a brand logo or font as a control signal.

3. Implement Post-Generation Verification

Apply OCR or a dedicated text detection model to verify the brand name appears correctly. If not, trigger a fallback: regenerate, or overlay the brand name using a deterministic method.

4. Design Fallback and Human-in-the-Loop

For critical ads, route low-confidence outputs to human review or automatically composite the brand name. This ensures correctness at the cost of some automation.

5. Define Metrics and Iterate

Measure success via brand name accuracy (e.g., exact match rate), visual quality, and latency. Use A/B testing to compare approaches and continuously improve.

Key Points to Mention

  • Text-to-image models (e.g., Stable Diffusion, DALL-E) struggle with rendering text accurately; conditioning on text embeddings alone is insufficient.
  • Fine-tuning on brand-specific images can improve generation of the brand name and style.
  • Post-generation OCR (e.g., Tesseract, Amazon Textract) can verify text correctness and trigger fallbacks.
  • Fallback strategies: regenerate with different seeds, use inpainting to correct text, or overlay the brand name programmatically.
  • Trade-offs: perfect accuracy may require human review or deterministic overlays, impacting scalability and cost.
  • Metrics: brand name accuracy, visual quality (e.g., FID), latency, and cost per image.

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

Q4

How would you design an A/B test for a ranking change that avoids contamination from network effects?

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

Said SUTVA violation out loud and the interviewer visibly perked up, so that's worth knowing.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by acknowledging that network effects can cause interference between treatment and control groups, leading to biased estimates. Then propose a design that isolates the ranking change, such as cluster randomization or switchback experiments, and outline how to measure and mitigate contamination. Finally, discuss validation and sensitivity analyses to ensure robustness.

Pro tip: Emphasize that in two-sided markets like Amazon, network effects often spill over across units; using a cluster-based design with a sufficiently large number of clusters and pre-experiment covariate balance checks can make your results credible to stakeholders.

1. Identify potential network effects

Map out how the ranking change could affect interactions between users, sellers, or products, and determine the likely pathways of contamination (e.g., shared inventory, social influence).

2. Choose a randomization unit that minimizes interference

Select a unit of randomization (e.g., geographic region, seller cluster, time-based switchback) that aligns with the network structure to reduce spillover between treatment and control.

3. Design the experiment with appropriate power and duration

Calculate sample size accounting for intra-cluster correlation, and set a duration that captures the ranking change's effect while avoiding novelty or primacy effects.

4. Measure and adjust for contamination

Use techniques like difference-in-differences, instrumental variables, or exposure modeling to quantify and correct for any residual interference.

5. Validate with sensitivity analyses and robustness checks

Conduct placebo tests, vary cluster definitions, and compare with alternative designs (e.g., switchback) to ensure conclusions are not driven by contamination.

Key Points to Mention

  • Cluster randomization (e.g., by geography or seller) to contain spillover
  • Switchback experiments for time-based network effects
  • Interference and spillover effects in two-sided markets
  • Intra-cluster correlation and adjusted sample size calculation
  • Difference-in-differences or instrumental variables to correct for contamination
  • Pre-experiment balance checks and placebo tests for validation

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

Q5

How do you handle position bias when training a ranker on implicit user feedback?

System DesignTechnical Trade-offs
Author's notes

Short answer: you have to mention it or they'll bring it up themselves.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining position bias and its impact on implicit feedback, then outline a multi-pronged strategy that combines model-based correction (e.g., inverse propensity scoring) with unbiased evaluation. Emphasize the trade-offs between different debiasing techniques and how you would validate their effectiveness in an online setting.

Pro tip: Highlight that position bias is not just a modeling problem but also a data collection issue; propose logging randomized exploration data or using counterfactual evaluation to ground your solution in real-world constraints.

1. Define the problem and its impact

Explain what position bias is in ranking, why implicit feedback (clicks, purchases) is biased by position, and how it can degrade ranking quality if ignored.

2. Choose a debiasing approach

Discuss methods such as inverse propensity scoring (IPS), click models, or unbiased learning-to-rank, and justify your choice based on data availability and business constraints.

3. Implement and integrate

Describe how you would implement the chosen method in the training pipeline, including propensity estimation, loss function modification, and handling of sparse data.

4. Evaluate and iterate

Outline offline evaluation using counterfactual or unbiased metrics, and online A/B testing to measure real-world impact, with a plan for continuous monitoring and refinement.

5. Address trade-offs and limitations

Acknowledge potential drawbacks such as increased variance, computational cost, or reliance on propensity model accuracy, and propose mitigation strategies.

Key Points to Mention

  • Inverse propensity scoring (IPS) and its variants (e.g., clipped IPS, self-normalized IPS)
  • Click models (e.g., examination hypothesis, cascade model) for propensity estimation
  • Unbiased evaluation metrics like counterfactual recall or IPS-weighted NDCG
  • Exploration data collection (e.g., epsilon-greedy) to reduce bias in logged feedback
  • Trade-offs between bias correction and variance, and between offline and online performance
  • Amazon-specific context: large-scale ranking, customer obsession, and long-term value

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