no oa, first round, and then invited to final round day. first round was 2 easy LC and 1 conceptual, Final rounds was behavioral, theory on AI/ML, 1 LC easy
Interview questions [1]
Question 1
alot on resume and theory on concepts covered in resume
The interview process started with a short 30-minute introductory interview, where we mainly discussed my previous experience, background, and motivation for the role. After that, I had three interviews in one day. One interview focused more on my previous projects and experience, while the other two were more technical and focused on machine learning theory, practical ML problem-solving, and model selection. The interviewers were very nice, professional, and the atmosphere was comfortable throughout the process.
Interview questions [2]
Question 1
Describe how you would build a model that recognizes whether a person is wearing glasses in a photo.
Two rounds of interview, both technical, in both rounds I was asked about my projects and very deep insights related to those, also the questions were very practical based ml questions, i was asked to do live coding also in both the rounds
I applied online. I interviewed at Microsoft in Mar 2026
Interview
3 interviews. One ML breadth mostly focused on Transformers. What are hallucinations, and how to avoid them, what is catastrophic forgetting, and similar questions. Another manager round with some HR questions, like sharing past learning experiences, was also asked about an open-ended ML system design for next-word prediction in an app. Third was the coding round, where they asked me to find an accurate threshold for classifying using a really bad classifier. The answer should be the F1 score, and I was supposed to code it up live.
Interview questions [1]
Question 1
What are hallucinations, and how to avoid them, what is catastrophic forgetting, and similar questions. Another manager round with some HR questions, like sharing past learning experiences, was also asked about an open-ended ML system design for next-word prediction in an app. Third was the coding round, where they asked me to find an accurate threshold for classifying using a really bad classifier. The answer should be the F1 score, and I was supposed to code it up live.
