Cleared the Nike ML engineer loop last quarter. Sharing this because the role is genuinely interesting and very different from what people expect when they think "nike ml job."
First: Nike has substantial ML infrastructure. I applied to the Consumer Intelligence team, which does personalization for Nike.com, NRC (Nike Run Club), and their digital commerce stack. This isn't a research job. It's applied ML engineering -- building and deploying models at scale, maintaining feature pipelines, working with experimentation frameworks. If you're coming from a research background, that's fine, but they're hiring for product impact, not papers.
The process: Recruiter call. Pretty quick, maybe 20 minutes. They asked about my domain (I came from recsys at a previous company, which was directly relevant). ML technical screen. 90 minutes, two parts. First part was an ML problem: you have a dataset of user activity on Nike.com (page views, product interactions, purchases). Design a model to predict 7-day purchase probability. They wanted to hear about feature engineering, how you'd handle class imbalance, what offline metric you'd use and why, and what online metric you'd track in production. Second part was a coding question: implement a simple gradient descent from scratch in Python. Not hard but you had to get it right. Full loop: four rounds.
Round 1: ML system design. Design the recommendation system for the SNKRS app. I talked through candidate generation, ranking, feature store considerations, and how you'd handle cold start for new shoe releases. This is the round that separates ML engineers from data scientists. You need to know how models get served, latency constraints, feature freshness tradeoffs.
Round 2: Applied ML depth. A senior MLE went deep on one project from my resume. They picked my recsys work and asked about every design decision. Why two-tower over BPR for this dataset. How did I handle position bias. What happened when I tried to retrain frequency X and what tradeoff I accepted. Bring your receipts.
Round 3: Coding + ML fundamentals. More Python. One DP problem (medium LC), one ML fundamentals question (they asked me to derive logistic regression gradient). The second part is easy if you've done it recently, embarrassing if you haven't touched it in a year.
Round 4: Behavioral + cross-functional. With a PM and a data scientist. Questions about working with non-ML stakeholders, communicating model limitations, shipping something you knew wasn't perfect.
Offer was around 180-195k base in Beaverton. I felt it was on the lower side of market for applied MLE work, especially given the scope of the recsys problem they're working on. But the ML problems are genuinely interesting if you care about consumer personalization at scale.