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

SeniorPrefer not to say

Apr 2026

Summary

Meta ML engineer interview with a system design question focused on Facebook Marketplace. Not much to go on from the session itself, but the question was meaty enough that I left wishing I'd prepped the ML side of marketplace ranking more carefully.

Questions Asked (1)

Design Facebook Marketplace from an ML perspective.

System DesignProduct StrategyTechnical Trade-offs

Author's notes

I went straight into listing ranking and relevance, which felt right, but I fumbled when they pushed on cold start for new sellers.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the core ML-driven problems in Facebook Marketplace: matching buyers with relevant listings, ranking search results, recommending items, and detecting fraud. Then outline an end-to-end ML system covering data, features, models, and evaluation, while discussing trade-offs between relevance, engagement, and trust. Finally, tie your design to Meta's scale and product goals, emphasizing iterative experimentation and metrics.

Pro tip: Anchor your answer in Meta's key metrics: meaningful social interactions and long-term user value, not just clicks. Show you understand that Marketplace is a two-sided platform where ML must balance buyer relevance, seller success, and marketplace health.

1. Clarify product goals and ML use cases

Ask clarifying questions about scope (e.g., search, recommendations, fraud) and define success metrics like CTR, conversion, and user retention. Identify the main ML problems: ranking, retrieval, personalization, and integrity.

2. Design data and feature pipeline

Describe data sources: user interactions, listing attributes, seller history, and social graph. Outline feature engineering for users, items, and context, including embeddings and real-time features.

3. Propose model architecture

Explain a two-stage system: candidate generation (e.g., two-tower embeddings) and ranking (e.g., deep learning with multi-task objectives). Include specialized models for fraud detection and pricing suggestions.

4. Define evaluation and iteration strategy

Cover offline metrics (AUC, NDCG) and online A/B testing with guardrail metrics. Discuss how to handle cold-start, feedback loops, and bias.

5. Address trade-offs and scalability

Discuss trade-offs between relevance and diversity, latency and model complexity, and personalization vs. privacy. Highlight infrastructure for large-scale training and serving.

Key Points to Mention

Two-stage ranking: candidate generation with two-tower models and ranking with deep neural networks.
Multi-task learning to optimize for multiple objectives like clicks, purchases, and user satisfaction.
Fraud detection and trust & safety models to prevent scams and counterfeit items.
Cold-start problem for new users and listings, and solutions like content-based features and exploration.
Evaluation metrics: offline (NDCG, recall) and online (CTR, conversion, retention) with A/B testing.
Scalability considerations: distributed training, real-time inference, and feature stores.

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