First was a standard recruiter screen where they mostly asked about my background and ML experience. After that I had a 45-min technical screen on coderpad - felt like a speed test - two LC medium problems with a pretty tight timer. Nothing crazy conceptually but you need to write clean code fast and keep explaining what you’re doing. The onsite was five round loop: two coding, one sys design, one ML design, and one behavioral. The ML design round was probably the toughest. They asked me to design something like a large-scale recommendation system for news feed and went pretty deep into data, features, models, and evaluation. The sys design round was more traditional infra stuff (scaling, distributed systems etc). Behavioral was again standard mostly past experience and handling ambiguity. One thing I realized though-they want specific signals and features in depth, not vague answers. I found a good Meta ML coach on Prepfully for the mock and he helped with ML design part prep. I’d say get a mock be it platform but get it; that helps a lottt more than we think!