← Google Interview Insights

Google·Machine Learning Engineer·Technical Phone Screen·Intermediate

IntermediatePrefer not to say
May 2026

Summary

Google ML interview, one question about batch vs real-time prediction tradeoffs. Short and focused, felt more like a screen than a deep dive.

Questions Asked (1)

Q1

Give an example of a use case that suits batch processing, and another that requires real-time predictions.

System DesignTechnical Trade-offsProduct Analytics & Metrics
Author's notes

I went with product recommendations for batch (run nightly, serve from a cache) and fraud detection for real-time.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Define batch and real-time processing, then provide a concrete use case for each, explaining why the latency and throughput requirements make one suitable over the other. Highlight the trade-offs and how the choice impacts system design and business metrics.

Pro tip: Tie each use case to a business metric (e.g., cost savings for batch, user engagement for real-time) to show product thinking, and mention that some systems use a hybrid approach (lambda architecture) to balance both.

1. Define batch and real-time processing

Briefly explain that batch processing handles large volumes of data offline with high throughput and latency tolerance, while real-time processing requires low-latency predictions on streaming data.

2. Provide a batch processing use case

Give a specific example, such as nightly recommendation model training or daily fraud detection report generation, and explain why latency is not critical and how batch efficiency reduces cost.

3. Provide a real-time prediction use case

Give a specific example, such as real-time ad targeting or instant credit card fraud detection, and explain why sub-second latency is essential for user experience or security.

4. Discuss trade-offs and design implications

Compare the two in terms of infrastructure (e.g., Hadoop/Spark vs. Kafka/Flink), cost, complexity, and freshness of data, and mention how the choice affects model updating and monitoring.

5. Conclude with business impact

Summarize how each approach aligns with different business needs and metrics, and optionally mention hybrid architectures like lambda or kappa for combined requirements.

Key Points to Mention

  • Latency requirements: batch tolerates hours, real-time needs milliseconds to seconds
  • Throughput and data volume: batch handles large-scale historical data, real-time handles continuous streams
  • Cost efficiency: batch is cheaper per record, real-time requires more resources
  • Use case examples: batch for periodic reporting, real-time for dynamic personalization
  • Infrastructure differences: batch uses distributed storage/compute, real-time uses stream processing
  • Hybrid approaches: lambda architecture for combining batch and real-time

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