Design and Model a Personalized Feed Recommender
Company: Shopify
Role: Machine Learning Engineer
Category: Machine Learning
Difficulty: medium
Interview Round: Onsite
Design a feed recommendation system and explain both its serving architecture and modeling choices. Connect the product objective, candidate generation, ranking model, training data, and evaluation.
### Clarifying Questions
The report gives the feed-recommendation prompt but no item type, traffic target, or objective. Clarify what appears in the feed, what a useful session means, eligibility and freshness rules, and the interaction data available. State any chosen baseline rather than assuming a particular company's existing architecture.
### What a Strong Answer Covers
A coherent retrieval-and-ranking pipeline, cold start, leakage-safe examples and negatives, loss selection, serving features, feedback bias, and offline/online evaluation. Keep model design connected to system constraints.
### Follow-up Questions
- How would new users and new items enter the feed before interaction history exists?
- What happens if engagement improves while users repeatedly hide recommended content?
Overview: Design a personalized feed from candidate retrieval to ranking and evaluation. Connect model losses, training labels, cold start, feature freshness, and feedback bias to the serving architecture.
