Applied on line. Took five weeks before the HR first contacted. The HR replied email timely since then. It was sort of weird as I didn't apply to a speech team, yet the HR mentioned the speech team would want to interview. Lacking first hand speech background. I used nlp courses from Stanford (the one on coursera and the one on deep leaning) and automatic speech recognition course offered by U. Edinburgh to gain some knowledge. Previous interview questions at glassdoor helped a lot in pointing to the directions. The interview was moved ahead for an hour the day before the scheduled interview. The interviewer was also late for 10 minutes. The interviewer start by asking what kind of machine learning techniques I have used in either my academic or internship career. I tried to link my experience to GMM, HMM, dimensional reduction, which shows up in automatic speech recognition. He then sort of based on what I said to ask questions. The questions were: 1. How does GMM/HMM work 2. Name some dimensional reduction method; I said PCA and we talked a bit about how PCA works and what's the physical intuiation 3. How K-means work, what kind of distance metric would you choose, what if different features have different dynamic range 4. How GMM works (EM algorithm) In the near future I suspect people would need knowledge in DNN/HMM. We then moved on to online coding, the HR had sent a link beforehand. I was asked to either code in C or Java. I screwed up, both in algorithm and in coding. The question was, given a function with inputs --an array with N randomly sorted numbers, and an int K, return output in an array with the K largest numbers. Overall it was a positive interview experience as the interviewer was nice in general.

Applied for Amazon AGI. After first round, it will go into full round of multiple interviews. Lots of modern LLM training technic questions. There are still some behavioral questions, but less than general Amazon roles.