Amazon

I got the opportunity to interview for the Applied Scientist Intern role at Amazon through the ML Summer School route. The process was quite different from a standard SDE internship because I knew from the beginning that the interviews would test both programming and machine learning. The first major filter was the online assessment. I received two DSA problems on HackerRank, and both were around the easy-to-medium level. I was able to solve them within the allotted time. The questions were not heavily ML-oriented, which initially surprised me, but I treated the OA like a normal coding assessment and concentrated on correctness, complexity, and edge cases. Recent candidates coming through the ML Summer School route have reported a similar two-question DSA assessment. After clearing the OA, I had the first technical interview. The interviewer started with a short discussion about my projects and then moved into DSA. One of the questions was a graph problem where all edges had equal weight. I recognized that BFS would give the shortest path and explained why before implementing it. The second question involved a Binary Search Tree. I was asked to reason about the operations rather than simply relying on a library implementation, so I walked through the recursive logic and its complexity. What I found useful was that the interviewer kept asking follow-ups. Instead of accepting the first solution, they wanted to know whether I could handle edge cases and what would happen if the constraints became larger. The round felt very similar to an SDE technical interview despite the Applied Scientist title. This DSA-heavy first round is also reported by recent Applied Scientist candidates. The second interview was completely different. It was the ML Depth and Breadth round, and this was the part I had prepared for most carefully. We started with my ML projects. I had to explain the problem I was solving, how I prepared the data, why I selected a particular model, which metrics I used, and what I would change if the results were poor. The interviewer then moved into ML fundamentals. We discussed decision trees, logistic regression, model evaluation, overfitting and underfitting, and deep-learning concepts. The questions were not simply “define this algorithm.” I had to explain why a particular technique worked and when I would choose one model over another. The discussion also went into neural networks and modern deep learning. I was asked about Transformers and how they differ from more traditional sequential architectures. The interviewer was particularly interested in whether I understood the underlying reasoning rather than whether I could reproduce textbook definitions. Recent Applied Scientist reports similarly mention Transformers, RNNs, decision trees, optimization, probability/statistics and project-level ML discussions. I performed much better in the ML round because I had prepared my projects in depth and could explain the decisions behind them. After completing the interviews, I received the selection confirmation. The biggest lesson for me was that the Applied Scientist interview required two completely different preparation tracks. DSA was necessary to get through the first technical stage, but the final ML round required genuine understanding of the mathematics, algorithms and practical decisions behind machine-learning systems.