The ML system design interview is where most candidates fall apart.

Not because they can't design systems — because they design the WRONG systems.

Here's what I mean: the interviewer says "design a recommendation system for an e-commerce platform." The candidate immediately starts talking about collaborative filtering algorithms and embedding spaces.

Wrong starting point.

The right starting point: "How many users? How many products? What's the latency requirement? What data do we have? What are we optimizing for — clicks, purchases, or long-term engagement?"

These questions shape the entire architecture. A recommendation system for 1,000 users is fundamentally different from one for 100 million. Real-time recommendations need a completely different architecture than daily batch recommendations.

The framework I use: requirements first (clarify scale, latency, data). Then high-level architecture (what are the major components?). Then data flow (how does data move through the system?). Then dive into the ML components. Then discuss tradeoffs, failure modes, and monitoring.

Interviewers don't want to see you recite an algorithm. They want to see you THINK like an engineer who's built real systems.

Practice this framework on 5-6 common problems (recommendations, search ranking, fraud detection, content moderation, forecasting, feed ranking) and you'll be ready for any ML system design question.