← Meta Interview Insights

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

SeniorPrefer not to say
Apr 2026

Summary

Meta MLE interview focused on designing a short-video recommendation system for Reels. The session was heavier on metric design and A/B testing than I expected, less about novel architecture and more about whether you actually understand how to evaluate what you build.

Questions Asked (6)

Q1

Design a personalized short-video recommendation system similar to Reels.

System DesignTechnical Trade-offs
Author's notes

The four-stage pipeline (retrieval, pre-rank, rank, re-rank) is basically assumed knowledge going in.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the problem scope and requirements, then outline the high-level system architecture covering data collection, candidate generation, ranking, and serving. Dive into the ML pipeline, discussing feature engineering, model choices, and evaluation metrics, while addressing trade-offs like latency vs. accuracy and scalability.

Pro tip: Emphasize the importance of real-time signals and user feedback loops, as Meta's systems heavily rely on fresh data to personalize content. Also, proactively discuss potential biases and mitigation strategies to show awareness of responsible AI.

1. Clarify Requirements and Scope

Ask questions to understand scale, latency constraints, business goals, and available data. Define success metrics like engagement, retention, and diversity.

2. High-Level Architecture

Sketch the end-to-end system: data ingestion, feature store, candidate generation, ranking, and serving. Mention offline vs. online components.

3. ML Pipeline Deep Dive

Detail candidate generation (e.g., two-tower models), ranking (e.g., deep learning with multi-task objectives), and feature engineering (user, video, context).

4. Evaluation and Iteration

Explain offline metrics (AUC, recall@k) and online A/B testing. Discuss how to handle cold start and feedback loops.

5. Trade-offs and Scalability

Address trade-offs: latency vs. model complexity, exploration vs. exploitation, and scalability challenges (e.g., distributed training, serving at scale).

Key Points to Mention

  • Two-stage architecture: candidate generation and ranking
  • Feature engineering: user embeddings, video embeddings, contextual features
  • Model choices: two-tower for retrieval, deep neural networks for ranking
  • Evaluation metrics: offline (recall@k, NDCG) and online (CTR, watch time)
  • Real-time personalization and feedback loops
  • Scalability and latency considerations (e.g., ANN search, model serving)

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

Q2

How would you define and structure the metrics for this recommendation system, from the north-star down to model objectives?

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

This is where things got real.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the product goal and user value, then define a north-star metric that captures long-term satisfaction. Work backward to structure a hierarchy of metrics: top-level health metrics, diagnostic metrics, and model-level objectives. Emphasize how offline model metrics connect to online A/B test outcomes and business impact.

Pro tip: Show that you understand trade-offs: e.g., optimizing for short-term engagement can hurt long-term retention, so include guardrail metrics. Also, mention that model objectives should be aligned with the north-star but may need proxies due to delayed feedback.

1. Clarify product goal and user value

Understand what the recommendation system is trying to achieve for users and the business. This sets the foundation for all metrics.

2. Define north-star metric

Choose a single metric that best captures long-term user satisfaction and business success, such as daily active users or long-term retention.

3. Build a metric hierarchy

Break down the north-star into sub-metrics (e.g., engagement, quality, diversity) and further into diagnostic metrics that can be directly influenced by the model.

4. Set model objectives and guardrails

Translate diagnostic metrics into model training objectives (e.g., CTR, watch time) and define guardrail metrics to prevent negative side effects.

5. Connect offline to online and iterate

Explain how offline model metrics predict online A/B test results, and how you would validate and iterate on the metric structure.

Key Points to Mention

  • North-star metric selection criteria (e.g., aligns with long-term goals, actionable, sensitive to changes)
  • Metric hierarchy: north-star, top-level health metrics, diagnostic metrics, model objectives
  • Examples of model objectives: CTR, watch time, revenue, diversity, freshness
  • Guardrail metrics to monitor unintended consequences (e.g., user churn, satisfaction surveys)
  • Offline-online correlation and A/B testing methodology
  • Trade-offs between short-term engagement and long-term retention

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

Q3

Walk me through how you would design an A/B test for a change to the ranking model.

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

My session apparently leaned hard into this.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the goal and defining a clear hypothesis for the ranking model change. Then outline a structured A/B test plan covering randomization, metrics, sample size, and analysis, while addressing potential pitfalls like network effects and novelty effects.

Pro tip: Emphasize guardrail metrics and long-term holdout to ensure the change doesn't harm user experience or long-term engagement, and discuss how you'd handle interference in social networks.

1. Define Hypothesis and Success Metrics

Clearly state the change and the expected impact, then select primary, secondary, and guardrail metrics (e.g., CTR, engagement, user satisfaction).

2. Design Experiment

Choose randomization unit (user, session), determine sample size and duration via power analysis, and set up control and treatment groups.

3. Address Potential Biases

Mitigate novelty effects, primacy effects, and network interference by using techniques like long-term holdouts, cluster randomization, or switchback tests.

4. Run and Monitor

Launch the experiment, monitor for technical issues and guardrail violations, and ensure data quality.

5. Analyze and Decide

Perform statistical analysis (e.g., t-test, bootstrapping), check for significance and practical impact, and decide whether to launch, iterate, or abandon.

Key Points to Mention

  • Randomization unit and assignment (e.g., user-level randomization to avoid contamination)
  • Primary and guardrail metrics (e.g., CTR, time spent, user reports)
  • Sample size calculation and power analysis to detect meaningful effect
  • Novelty and primacy effects and how to mitigate them (e.g., long-term holdout)
  • Network effects and interference in social networks (e.g., cluster randomization)
  • Statistical significance vs. practical significance and long-term impact

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

Q4

What features would you use, and how would you handle social graph signals?

System DesignTechnical Trade-offs
Author's notes

Went fine.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the product context (e.g., feed ranking, friend recommendations, ads) and the scale of the social graph. Then outline a feature engineering strategy that combines node-level, edge-level, and graph-level features, and discuss how to handle the unique challenges of social graphs like sparsity, dynamism, and privacy.

Pro tip: Emphasize that social graph signals are most powerful when combined with user-item interaction features, and always mention privacy and fairness considerations—Meta cares deeply about responsible AI.

1. Clarify the problem and constraints

Ask questions to understand the specific task (e.g., ranking, recommendation), the scale (billions of nodes/edges), latency requirements, and privacy constraints. This shows you can scope the problem before diving into solutions.

2. Identify relevant social graph signals

List potential signals such as degree, centrality, community membership, tie strength, common neighbors, and interaction frequency. Categorize them into node-level, edge-level, and neighborhood-level features.

3. Design feature engineering and representation

Explain how to compute these features efficiently (e.g., using graph embeddings like node2vec, GraphSAGE, or precomputed aggregates) and how to handle dynamic updates. Discuss trade-offs between batch and real-time computation.

4. Address challenges and trade-offs

Discuss sparsity (e.g., cold-start users), scalability (sampling, distributed processing), and privacy (differential privacy, federated learning). Mention how to evaluate feature importance and avoid leakage.

5. Integrate with the ML pipeline

Describe how these features feed into the model (e.g., as embeddings, raw features), how to handle missing values, and how to monitor and update them in production.

Key Points to Mention

  • Graph embeddings (e.g., GraphSAGE, node2vec) for scalable representation learning
  • Neighborhood aggregation and sampling techniques to handle large graphs
  • Tie strength and interaction frequency as edge features
  • Community detection and clustering coefficients for structural context
  • Privacy-preserving techniques like differential privacy and federated learning
  • Handling dynamic graphs with incremental updates and temporal features

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

Q5

How do you address position bias in training data, and how would you monitor for training-serving skew?

A/B Testing & ExperimentationRoot Cause Analysis
Author's notes

Inverse propensity weighting came up and I explained it reasonably well.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining position bias and training-serving skew, then outline a systematic approach to detect and mitigate position bias in training data, followed by a robust monitoring strategy for training-serving skew. Emphasize practical techniques like counterfactual logging, propensity scoring, and statistical tests, and tie them to real-world ML engineering practices.

Pro tip: Highlight the importance of logging serving-time features and predictions to enable continuous monitoring, and mention that at Meta, even small skews can significantly impact user experience due to scale. Also, discuss trade-offs between bias mitigation and model performance.

1. Define and Detect Position Bias

Explain what position bias is (e.g., users click top results more often regardless of relevance) and how to detect it using click models, propensity scores, or randomization.

2. Mitigate Position Bias in Training Data

Describe techniques such as inverse propensity scoring (IPS), counterfactual logging, or using unbiased estimators to reweight training examples.

3. Define and Detect Training-Serving Skew

Define skew as differences between training and serving data distributions or feature computations, and list common causes (e.g., feature pipeline inconsistencies, time-based drift).

4. Monitor for Training-Serving Skew

Outline a monitoring plan: log serving features and predictions, compare distributions with training data using statistical tests (e.g., KL divergence, PSI), and set up alerts for significant deviations.

5. Iterate and Validate

Emphasize continuous validation through A/B tests, shadow deployments, and feedback loops to ensure mitigation strategies remain effective and skew is addressed.

Key Points to Mention

  • Inverse Propensity Scoring (IPS) for debiasing
  • Counterfactual logging or randomized experiments to collect unbiased data
  • Feature consistency checks between training and serving pipelines
  • Statistical tests like KL divergence or Population Stability Index (PSI) for skew detection
  • Logging serving-time features and predictions for monitoring
  • A/B testing and shadow deployment to validate fixes

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

Q6

How would you handle cold start for new users and new videos?

System DesignProduct Sense & Ideation
Author's notes

Short answer: content-based fallback and exploration bandits.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the problem scope and success metrics, then propose a two-sided solution: one for new users (user cold start) and one for new videos (item cold start). For each, outline a progression from content-based/feature-based methods to exploration and rapid online learning, and discuss how to evaluate and iterate.

Pro tip: Emphasize that cold start is not just a modeling problem but a product and systems problem—tie your solution to business metrics like user engagement and retention, and mention how you'd balance exploration costs with long-term gains.

1. Clarify requirements and metrics

Ask about the product context (e.g., feed, stories, reels), what signals are available at cold start, and define success metrics (CTR, watch time, retention).

2. New user cold start strategy

Propose using onboarding signals, demographic/context features, and content-based recommendations; then transition to collaborative filtering as user data accumulates.

3. New video cold start strategy

Leverage content features (visual, audio, text) and creator metadata to match videos to users; use multi-armed bandits or Thompson sampling for exploration.

4. Unified exploration and learning system

Design a system that jointly handles both cold start types, using contextual bandits with shared embeddings and fast online updates to quickly learn from early interactions.

5. Evaluation and iteration

Define offline and online evaluation (A/B tests), monitor for biases, and plan for continuous improvement as more data arrives.

Key Points to Mention

  • Content-based filtering using metadata and multimodal features (text, image, audio) for new videos.
  • Demographic, contextual, and onboarding signals for new users.
  • Exploration-exploitation trade-off via multi-armed bandits or reinforcement learning.
  • Transfer learning and meta-learning to leverage data from similar users/videos.
  • Hybrid approaches that blend content-based and collaborative signals.
  • Online learning and fast model updates to adapt quickly.
  • Evaluation metrics: CTR, watch time, retention, and diversity.
  • Handling popularity bias and ensuring fairness in exploration.

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