The interview was an initial phone screen conducted on Amazon Chime (basically a chatroom, with no video or screenshare) and LiveCode.

• 15 minutes of behavioral questions
• 15 minutes of machine learning fundamentals
• Approximately 25 minutes for a coding question
• 5 minutes at the end for the candidate to ask questions

Behavioral questions were heavily oriented towards Amazon's leadership qualities:

• Name a time you were innovative.
• Name a time you delivered a simple solution to a complex problem.

Follow-up questions included how to quantify the level of success in projects brought up.

Machine learning fundamentals:

• How to deal with a troublesome dataset (interpretation open-ended, so think data cleaning, etc.).
• How to deal with misrepresentative training data (imbalanced dataset, overfitting, explain how L1/L2 regularization works at an optimization level).
• How to deal with a large dataset where only a few examples are labeled (semi-supervised learning).

The coding question was: https://leetcode.com/problems/find-original-array-from-doubled-array/