Amazon Senior Applied Scientist Interview Experience — Passed the Tech Screen Despite an Unfinished Graph Question

An Amazon recruiter found me on a job site and had me apply for a senior applied scientist role. The recruiter met with me first and walked me through the general shape of the tech screen — ML breadth plus coding.

1. ML Breadth. The interviewer was pretty nice. This part asked a lot of fundamental questions, like:

1. What is a p-value?
2. What is overfitting and underfitting, and how do you address them?
3. What is causal inference? What methods are there?
4. What is encoding and decoding?
5. Explain gradient descent and backpropagation.
6. Explain gradient vanishing/exploding, and how to mitigate it.
7. How do you handle highly imbalanced data?
8. Describe a scenario where you have 99% accuracy but the model performance is still bad — how would you tackle it?
9. What is an A/B test? If an A/B test shows an anomaly, what could be the possible causes?

2. Coding was a question from a LeetCode-style practice site. This part was only about 15 minutes, and to my surprise it turned out to be a hard-level question. I hadn't seen it before, and I didn't expect it to be a graph problem, so I asked the interviewer for a hint. He told me to just go with brute force. I roughly explained my approach, wrote maybe a third of the code, and then ran out of time.

Since I didn't finish the coding part, I thought I'd failed, but half an hour later the recruiter told me I passed. The recruiter said that for applied scientist, the bar for ML is higher and needs to be raised, but the coding bar only needs to be at the SDE 1 level. The recruiter read me the interviewer's comments — ML was very strong, and in coding I communicated my approach well. So even if you don't finish writing the code, don't get discouraged — staying in communication with the interviewer and clearly explaining your approach really matters. Thanks to the interviewer for not making it hard on me.