The phone screen was surprisingly short, only about 20 minutes, focusing on my background and some general ML concepts. Then came the technical round where I had to design a ranking model for Instacart’s search results. Midway through, it hit me that I had just practiced a very similar problem on PracHub, which made me feel a bit more confident. However, the depth of the discussion, especially around metrics and cold-start items, was intense. In the end, I didn’t receive an offer, but it was a decent experience overall.

Design a ranking model for Instacart's search results. Walk through your feature engineering, which includes item-side, query-side, user-side, and query-item interaction features. Discuss candidate generation versus ranking stage architecture. Explain how you'd train with implicit click feedback while controlling for position bias, offline evaluation metrics like NDCG versus business KPIs such as add-to-cart rate, and how you'd handle cold-start items and avoid overfitting to short-term engagement signals.