Amazon Machine Learning Engineer Interview Experience — 90 Minutes Covering System Design, RL, and a Graph-Cycle Coding Question

Basically felt like going into battle with my head already wrapped in bandages, wounded before I even started. It was originally scheduled for one hour, but it ended up running an hour and a half.

The coverage was extremely comprehensive:

1. First, a self-introduction covering the projects I'd done, but with a bit of a science-breadth angle mixed in — this took 3-5 minutes.

2. Right after that came a deep-dive system design discussion: model design, the objective function, how the model is trained, how it's used, and so on.

3. Then the topic switched to RL — they asked what the common basic components of RL are, and I rattled through them one by one off the top of my head.

4. After 40 minutes, it suddenly jumped into a coding question. The problem was something like: given a set of item edges, determine whether it's a valid embedded/nested structure. My first instinct was to reach for topological ordering — that's not wrong, but this problem can also be solved with a plain BFS. Even though it's a directed graph, all you need is to prove that no infinite loop (cycle) forms.

5. After I finished coding, they had me explain it again. By the time I was done, 75 minutes had already passed, and then they moved into Q&A, asking about where GenAI is being used internally, and so on.