← Airbnb Interview Insights

Airbnb·Machine Learning Engineer·Onsite - System Design / Architecture·Senior

SeniorPrefer not to say
Apr 2026

Summary

Airbnb ML system design round focused entirely on one deep question about photo selection for listings. It was a long session and they really wanted you to go end-to-end, not just sketch a model.

Questions Asked (1)

Q1

Design a machine learning system that automatically picks the best cover photo for an Airbnb listing to improve booking conversion. Walk through the full pipeline from problem framing to serving.

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

This one ate up the entire session.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business objective and defining a measurable target (e.g., booking conversion lift). Then outline an end-to-end ML pipeline: data collection, feature engineering, model training (e.g., pairwise ranking of photos), offline evaluation, online A/B testing, and serving with monitoring. Emphasize trade-offs between model complexity, latency, and business impact.

Pro tip: Frame the problem as a ranking task with implicit feedback (e.g., which photo was chosen by hosts or which led to bookings) and highlight the importance of counterfactual evaluation to avoid bias from historical photo selections.

1. Problem Framing & Metrics

Define the goal: increase booking conversion by selecting the best cover photo. Choose metrics: primary (booking conversion rate), secondary (click-through rate, photo engagement), and guardrail (host satisfaction, diversity).

2. Data Collection & Feature Engineering

Gather data: listing photos, metadata (resolution, brightness, composition), host and listing features, and user interactions (views, bookings). Engineer features from images (CNN embeddings, aesthetic scores) and context (seasonality, location).

3. Model Training & Offline Evaluation

Train a model to rank photos per listing. Use pairwise or listwise ranking losses (e.g., LambdaRank) with implicit feedback. Evaluate offline with NDCG, precision@k, and counterfactual estimators (IPS) to handle selection bias.

4. Online Experimentation & Deployment

Run A/B tests to measure causal impact on booking conversion. Deploy model via a serving system that scores photos in real-time or batch, with fallback to default. Monitor latency, model drift, and business metrics.

5. Iteration & Feedback Loop

Continuously collect new data, retrain models, and refine features. Incorporate host feedback and explore multi-objective optimization (e.g., conversion vs. diversity).

Key Points to Mention

  • Use of pairwise ranking models (e.g., LambdaRank) for photo selection
  • Handling implicit feedback and selection bias via counterfactual evaluation
  • Feature engineering from images (CNN embeddings, aesthetic scores) and context
  • A/B testing framework with proper randomization and guardrail metrics
  • Serving architecture with low latency and fallback mechanisms
  • Trade-offs between model complexity, interpretability, and business impact

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