Test your knowledge of production ML systems with 16 interview-style multiple choice questions covering recommendation systems, LLM serving, and more.
Test your knowledge with 16 questions
Correct
In a recommendation system, what is the primary purpose of the Two-Tower architecture?
What is the main purpose of a Feature Store in ML systems?
In cascade ranking for search systems, why do we use multiple stages?
What is 'training-serving skew' in ML systems?
In a fraud detection system, what is the typical trade-off when tuning the decision threshold?
What is the purpose of 'shadow deployment' in ML systems?
In LLM serving, what is 'PagedAttention' used for?
What is the main advantage of using HNSW (Hierarchical Navigable Small World) for vector search?
In a news feed ranking system, what is 'position bias' and how is it typically addressed?
What is the purpose of 'model distillation' in ML serving?
In ads CTR prediction, why is 'calibration' important?
What is 'concept drift' in production ML systems?
In hybrid search systems, what is 'Reciprocal Rank Fusion (RRF)'?
What is the purpose of 'continuous batching' in LLM serving?
In ML system design, what is the 'cold start problem' for recommendations?
What is 'A/B testing interleaving' and why is it preferred over traditional A/B testing for ranking systems?
Inspired by ML systems at Google, Meta, Netflix, and other tech giants. Content is for educational purposes and interview preparation.