Amazon Research Scientist, Machine Learning interview questions

Recruiter emailed me to set up appointment. I gave a number of blocks of time when I was available to do the interview. I didn't hear back from them and, in the mean time, a meeting was scheduled at work resulting in one of the blocks of time to no longer be available. I contacted the recruiter to inform her that one of the blocks of time were no longer available. The next day, they schedule the phone interview for precisely when my meeting occurred. I contacted and phoned the recruiter to ask if I could push the phone interview later because of a meeting that was scheduled that I couldn't miss at work. Time passed and the recruiter never contacted me. Another recruiter/HR person called me, not at the phone number I had given, but on my cell phone to tell me that they can push the phone interview later. I was somewhat surprised that a company that prides itself on logistics would be a tad bit sloppy with the logistics of setting up the phone interview. Nonetheless, the interviewer explained the structure of the interview, they ask open ended questions and I would answer them. They gave a general idea of the areas on which the questions would be asked. The interviewer spoke really really really fast and came across as trying to artificially create a pressure filled environment, which came across as very weird. The first question concerned a linked-list data structures question. It seemed strange since the position was a research position and not a software-engineering position. Easy enough, but I was still taken off guard as I was hoping for a Machine Learning question as that is my background and that is what the position concerned. I explained how one would implement a sorting algorithm, then the two alternatives for finding median based on whether you had an even or odd number of items in the linked list. They actually wanted a description of what the code would do, so I explained you implement a while loop iterating as many times as described by the indices computed for obtaining the median. The "how would you actually implement it" was a strange question as it isn't really the focus of what a researcher will solve, at least not in Machine Learning. Then the topic turned to a general design question. They asked where I would start if I were designing a robot lawn mower. I began with what I would have my team do. They interrupted me and said, what if you didn't have a team and it was just you. This process of frequently interrupting me and changing the scope or scenario occurred throughout my interview. After a point, I had lost interest in the job during the interview. Anyway, I began with requirements and how I would actually mow lawns and talk to landscapers who mow lawns to understand the environment, variations, types of equipment and issues encountered with mowing laws; that is understand the domain. Again, I was interrupted and the scope shifted to assume someone had solved all the mechanical problems and implemented the hardware platform. They were asked where would you start. OK, so I began with the sensors I would use (Camera, Lidar, etc.) and before I could start to explain sensor fusion and localization, I was interrupted and the scope was changed to cameras only. So, I explained the types of features you might extract from the camera in addition to a system of beacons to make the problem easier. They interrupted again to say what if you didn't have beacons and only cameras. OK, so I described the issues with both navigation and detecting the boundaries of the lawn to be mowed in addition to detection of objects. They interrupted me again asking where I would put the camera. I described multiple cameras, one forward facing and another with a hemispherical panoramic lens. As I continued to describe the system, they interrupted me again asking how I would recognize people. I explained you are concerned about living things in general and that a thermal camera could be added to detect warm bodies in addition to motion queues. They interrupted me again scoping the problem based on static images. I described a skeletal model. They asked my how I would literally implement it. I described at a high level a parts based model because the person might be partially occluded. They asked specifically how I would implement the parts based model. Then they changed the topic to give an example where I was on a project that did not work out as planned. I described this and they asked who made the decision. Then they asked me to give an example of a project where I had multiple alternatives from which to choose for components and what choice I made. Then they asked me if I had any questions. I asked how about a Machine Learning question? They didn't really have an answer only that the allotted time for my interview was done.