Topics: Machine Learning, System Design, Behavioral Questions, Recommender Systems, Model Serving
Location: San Francisco Bay Area
Interview date: 2025-02-15
Got offer: False
Summary
Round 1: ML System Design
Question: Design a recommender system: Where to save data? How to serve models? What is a reasonable number of candidates to pass into the ranking model, and how is it calculated? Given a 60GB memory GPU, how to calculate the batch size in model serving?
Round 2: Behavioral (BQ)
Question: What is the 2-year plan for your current recommender system project? What are the most important qualifications you value when hiring for your team? What kind of candidate would you hire to fill the biggest gap in your current domain?
Details
ML System Design Questions and Answers
Building a Recommender System
Data Storage: How should data be stored?
Model Serving: How should models be served efficiently?
Candidate Selection:
How many candidates should be passed into the ranking model?
The answer should be in the range of a few thousands.
Calculation factors:
Model inference complexity
Distributed inference cost
Number of instances of each cluster in the distributed system
Batch Size Calculation: Given a 60GB memory GPU, how should the batch size be calculated in model serving?
Behavioral Questions
2-year plan for your current recommender system project.
Most important qualifications for new hires on your team.
What kind of candidate would you hire to complete the biggest gap in your current domain?