I recently interviewed for Staff AI Eng at LinkedIn. My experience with the interviewers across the rounds was pretty good, but things were busy at home and I didn't have the energy to prepare much. In the end they downleveled me to Senior and I didn't take it.
There were five rounds in total, one hour each.
Staff AI Modeling & Fundamentals: about the same as what you see posted on the forum. They ask the standard memorize-it fundamentals: Transformer complexity, how to optimize inference, and they also asked about GQ Attention and Flash Attention. I hadn't prepared much on these, so after that the interviewer just asked about things on my resume. You really should prepare more fundamentals for this round.
Staff AI System Design: a pretty typical ML problem, how to increase user time spent on a feed ranking system.
Staff Coding - Data: one question was a dice roll. Given the probability of each side, implement the dice roll function, with some follow-ups on corner cases. Then they asked another question that can be turned into overlapping interval merge.
Host Leader - AI: a behavioral interview. The interviewer gave a case study problem. I thought it was meant to be a technical discussion, so I started talking about technical stuff right away. Only later, with some guidance, did I realize they actually wanted to get at behavioral signal through a concrete problem. I went off track a bit.
Staff Coding with AI - AI: the IDE comes with various coding agents, and you get a repo with a few files. There were four or five sub-questions in total, and I only got halfway through the third one. You need to be careful with this one. I mostly use Claude Code, where I plan first and then execute, and that has a lot of overhead. The agent's output was long every time, and eventually I had no time left to read it. The problems weren't actually hard and I knew how to solve each one right away. I think it would have been better to use a simple model and solve things quickly.
Discussion
Loading comments…