I applied in-person. The process took 6 weeks. I interviewed at Amazon (Berlín) in Jan 2022
Interview
This was the first time I had such interview experience which to be honest was not the best. The first interview was rather easy with some pair-programming and some typical DS questions.
The final round was 6 session each around 1 hour. I had so many sessions all focusing on the leadership principles and they had overlapped questions "tell me about the time....".
There were few deep questions regarding Data Science and Machine Learning. There were only one session which I can call it deep dive which was so free style. The interviewee didn't have any deep follow up questions, I also asked if there is any tool so I can write some formulas or sketch the pipeline but there were no tools available ...
The interviewee also gave me a hint on the project to be retail-related (that was the team which I was applying for) although the question was `tell me about a project that you've work and have deep knowledge of`. I didn't have any recent retail-related project, of course, cause I was working in other industries in my past job and if it was a hard filter they should have notice it earlier in the resume review.
At the end of the day I had no idea what was the expectation, if I did good or not.
Overall, I would not recommend this type of process for any Data Science job interviews and all in all was a bad experience.
I applied through an employee referral. The process took 4 weeks. I interviewed at Amazon (Arlington, VA) in Dec 2021
Interview
5 interviewers, each session is 1 hour. Among the 5 interviewers, 3 do not have any DS background. So the questions are completely Leadership Principle (behavior questions). 2 have DS background, but each only asked me 1 or 2 quick questions related to DS. The remained questions are all behavior. I have no idea how they made the decision.
Interview questions [1]
Question 1
99% of the 5 hour interview are behavior questions
I applied online. The process took 4 weeks. I interviewed at Amazon in Mar 2021
Interview
Got an email inviting me to interview. It said there would be behavioral questions as well as anything from statistics and ML could be covered, as well as a coding section.
I prepared for a few weeks and then had the call with an applied scientist. The first half was spent going through projects on my resume. I think Amazon encourages interviewers to be "tough" by interrupting you, zeroing in on random technical details, and talking fast. I feel like I kept up ok, but it felt adversarial. The interviewer kept trying to lead me to say a certain phrase, even if I was describing it already - e.g. saying "moving average" instead of "recalculating the mean in a moving window."
The second half was the coding section. Again, it was difficult to understand what exactly the interviewer wanted - they seemed to make it up as they went - and they were very pushy and impatient. But the best came last - I wrote the Python code to train an ARIMA model and was asked to generate a forecast, but was then repeatedly told I was doing this step wrong, that I needed to pass in data to be able to generate a forecast. I was confused about this and disagreed but was steamrolled over. When I looked it up later, though, the interviewer was wrong - you only need to specify the number of steps you want in your forecast; you don't need to pass in additional data.
Third part was behavioral. No, just kidding. That part of the preparatory emails was wrong. Would have been nice to have that time to prepare more for what the interview actually covered.
Got an email rejection the next day. Overall pretty negative experience - I feel the interview was arbitrary and unnecessarily high-pressure, and I had no idea what the actual interview was going to cover until it started.
Interview questions [1]
Question 1
Describe the iterative steps k-means takes to find clusters, how would feature importances change if you retrained the model with a duplicate column, walk through the steps of loading data from S3, cleaning it, then training an ARIMA model and generating forecasts