Design a cold-start video recommender
Company: Voleon
Role: Machine Learning Engineer
Category: ML System Design
Difficulty: hard
Interview Round: Technical Screen
Design a cold-start recommendation pipeline for a short-video platform. The system must work for both new users with little or no interaction history and new videos with little or no engagement data. You may assume graph information is available, such as user-user, user-video, creator-video, hashtag-video, and video-video relations.
Describe:
- The offline and online architecture
- How you would use content features and graph features
- Candidate generation and ranking stages
- How to handle exploration versus exploitation
- Real-time feedback and model updates
- Metrics, guardrails, and experiments
- How the system transitions from cold-start recommendations to fully personalized recommendations
Quick Answer: This question evaluates proficiency in ML system design for recommender systems, focusing on cold-start handling, integration of content and graph features, candidate generation and ranking, exploration–exploitation trade-offs, real-time feedback and model updates, and experimental metrics and guardrails.