Meta·Data Scientist·Technical Phone Screen·Senior

Senior

Jul 2026

Summary

Meta DS interview focused on ML system design, specifically ranking posts for Newsfeed. The whole thing was one big open-ended problem and they clearly wanted to see how deep you could go on features, modeling choices, and evaluation.

Questions Asked (1)

Design a machine learning model to rank Facebook Groups posts in a user's Newsfeed. Walk through your feature choices, labels, model family, and how you'd evaluate and deploy it.

System DesignA/B Testing & ExperimentationTechnical Trade-offs

Author's notes

This is a beast of a question and I underestimated how much they'd push on each layer.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the problem: define the ranking objective (e.g., maximize meaningful interactions) and constraints (latency, fairness). Then walk through features, labels, model choice, evaluation, and deployment in a structured, iterative manner, emphasizing trade-offs and experimentation.

Pro tip: Emphasize that ranking is not just about prediction accuracy but about optimizing long-term user value; mention how you'd balance multiple objectives and guard against feedback loops.

1. Clarify Objective and Constraints

Define the ranking goal (e.g., maximize meaningful interactions) and constraints like latency, fairness, and scalability. Align with Meta's mission and business metrics.

2. Feature Engineering

Identify user, post, group, and context features. Include interaction features, temporal signals, and embeddings. Discuss how to handle sparse and high-cardinality features.

3. Label Definition and Model Selection

Choose labels (e.g., clicks, comments, shares, dwell time) and combine them into a single utility score. Select a model family (e.g., GBDT, DNN, or multi-task learning) suitable for large-scale ranking.

4. Evaluation and Offline Metrics

Use offline metrics like NDCG, MAP, or AUC, and design online A/B tests to measure impact on user engagement and business metrics. Consider counterfactual evaluation.

5. Deployment and Iteration

Describe a scalable serving architecture (e.g., model server, feature store) and monitoring. Plan for continuous retraining and experimentation to adapt to changing user behavior.

Key Points to Mention

Multi-task learning to predict multiple engagement types and combine them into a ranking score.
Handling position bias and feedback loops in training data (e.g., using inverse propensity scoring).
Feature importance and engineering: user-group affinity, post recency, content type, and social signals.
Trade-offs between model complexity and latency for real-time ranking.
A/B testing framework: define guardrail metrics, run power analysis, and measure long-term effects.
Fairness and diversity considerations to avoid filter bubbles and ensure equitable group exposure.

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