System Design for Data Scientists

Six years of ML design interviews. Both sides of the table. Hiring committees at FAANG and adjacent. A few hundred candidates watched, coached, debriefed.

Here is what I figured out. The candidates who get staff offers are not smarter. They know the shape of the answer. What to ask in the first ninety seconds. Which three tradeoffs to name out loud. How to size the system before drawing a box. When to bring up feedback loops without being asked. How to close in thirty seconds with build, sacrifice, watch.

That shape is teachable. Nobody teaches it. So I wrote the book.

22 chapters covering everything an ML design round actually tests. Problem formulation. Feature stores. Two-stage retrieval.

Embeddings. Recsys. Fraud. Forecasting. Pricing and marketplaces. Serving. Monitoring. A/B testing. Causal methods beyond A/B. RAG. Product sense. Plus four full case studies with real transcripts, pauses and reversals included, because that is how interviews actually sound.

Each chapter ends with a one-page cheat sheet you can photograph between rounds.

If you have a staff-track onsite in two weeks and you freeze when the interviewer says "design YouTube's recommendation system," this is for you. If you have a quarter, it doubles as a reference. Read straight through if you have the time. Read by topic if you don't.

This is the book I wish someone had handed me before my first onsite.