Meta·Machine Learning Engineer·Technical Phone Screen·Intermediate

IntermediatePrefer not to say

Apr 2026

Summary

Interviewed at Meta for a machine learning or data-related role, got asked a pretty foundational question about prediction serving. Short and to the point.

Questions Asked (1)

What is the difference between batch predictions and real-time predictions?

System DesignTechnical Trade-offs

Author's notes

Seems basic but I fumbled the tradeoffs a bit.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining both batch and real-time predictions clearly, then contrast them across key dimensions like latency, throughput, data freshness, and infrastructure. Finally, discuss trade-offs and when to choose each approach, ideally with examples relevant to Meta's scale.

Pro tip: Emphasize that the choice isn't binary—many production systems use a hybrid approach (e.g., batch for precomputation, real-time for dynamic features). Mentioning this shows you understand practical system design beyond textbook definitions.

1. Define batch predictions

Explain that batch predictions are generated periodically on large volumes of data, often offline, and stored for later use. They prioritize throughput over latency.

2. Define real-time predictions

Explain that real-time predictions are generated on-demand with low latency, often using streaming data and online models. They prioritize latency over throughput.

3. Compare across dimensions

Contrast them on latency, throughput, data freshness, cost, complexity, and use cases. Highlight that batch is suitable for non-urgent, large-scale scoring, while real-time is for interactive or time-sensitive applications.

4. Discuss trade-offs and hybrid approaches

Acknowledge that real-time systems are more complex and costly but provide fresher predictions. Mention hybrid architectures like lambda or kappa that combine both.

5. Relate to Meta's context

Give examples of where each might be used at Meta, such as batch for daily ad performance predictions and real-time for ranking feed content or detecting spam.

Key Points to Mention

Latency requirements: batch (minutes to hours) vs. real-time (milliseconds to seconds)
Data freshness: batch uses historical data, real-time uses latest data
Infrastructure: batch uses offline storage and scheduled jobs, real-time uses streaming and online serving
Cost and complexity: real-time is more expensive and complex to maintain
Use cases: batch for recommendations, reporting, or periodic scoring; real-time for fraud detection, dynamic pricing, or personalization
Hybrid approaches: precompute batch features and combine with real-time features for low-latency serving

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