Microsoft Data Scientist Interview Questions
& Process
Real candidates share what happened, how many rounds they had,
and how the experience turned out.
Based on 374 interview experiences · FREE TO READ
Candidate interview experiences
First-hand accounts from people who interviewed at Microsoft.
Microsoft
Applied Scientist
Intern Difficult Negative experience No offer 1 round
Interview process
Phone screen Technical screen
Interview formats
Behavioral Technical Coding
I recently had a first-round screening that was honestly a masterclass in how not to treat candidates. Even though I was selected for my background in Data Science and AI, the experience was a complete mismatch: Lack of basic courtesy: The interviewer kept their camera off the whole time while expecting me to stay fully engaged. Total role misalignment: Not a single question was about my resume, ML/Stats knowledge, or past projects. It was obvious the interviewer didn’t care about my background. I have no idea why they chose me then... Argumentative, not evaluative: When I was asked about my strengths, the interviewer responded by directly arguing against my answers instead of asking professional follow-ups. We went back and forth like five rounds. It felt like he just disliked my strengths. Technical chaos: Even though they gave a HackerRank link, I was told to code in a local IDE only based on verbal instructions, which made everything unnecessarily confusing and prone to miscommunication. It’s really disappointing to see a company show such a lack of respect for a candidate’s time and expertise.
Confirmed questions4 questions
- 1. What are your advantages and disadvantages for this position?
- 2. How did you compute the GPU memory usage for the 4B Qwen model?
- 3. How does LoRA work for optimization? More specifically, at the GPU level?
Microsoft
Applied Scientist
Lead Average Positive experience Accept offer 4 rounds
Interview process
Recruiter call Technical screen Onsite Offer
Interview formats
Behavioral Technical Presentation Other
I applied through a general posting labeled "Senior/Principal Applied Scientist" and because of that I went through the "new" hiring process for applied scientists. It was by far the best hiring experience I had this past year. After an intro call with HR I was matched with 2 senior Applied Scientist managers and a director, and I had both technical and behavioral interviews that went deep into my past research experiences and interests. After passing this step I was invited to give a research seminar on a topic of my choice, and apparently the entire division attended. At the end, I had a few shorter interviews with team leaders who had open positions and were interested in me, and I received an offer after those conversations.
Confirmed questions1 question
- They gave me a Microsoft problem and I had to answer in a free-form way how I would solve it and scale the solution into production.
Microsoft
Applied Scientist
Mid Average Positive experience Decline offer 3 rounds
Interview process
Recruiter call Take home Technical screen
Interview formats
Behavioral Technical Coding
1. Application and Resume Screening Candidate sends in a resume and cover letter. Recruiters or automated tools look over them for qualifications, experience, and how well they match the role. 2. Recruiter Phone Screen A brief (15–30 min) call with a recruiter. Purpose: Confirm basic qualifications, talk about career goals, describe the role, and get a sense of communication skills. Might also cover logistics like availability and salary expectations. 3. Technical/Skills Assessment (if applicable) For technical positions, you may be asked to: Do a coding challenge or a take-home task. Take online assessments on platforms like HackerRank, Codility, or SHL. For non-technical roles, this could be: A case study, writing exercise, or portfolio review.
Confirmed questions6 questions
- What is the difference between supervised and unsupervised learning? Can you mention some real-world examples for each type?
- Explain what overfitting and underfitting are. How can you avoid overfitting? What do you mean by the bias-variance trade-off? What are precision, recall, and F1-score, and in which situations would you pick one metric over another?
- Describe how a decision tree works. How is logistic regression different from linear regression? How does the K-nearest neighbors (KNN) algorithm operate? What is regularization, and how do L1 and L2 regularization differ?
Microsoft Data Scientist Interview Questions
Quoted word for word from Microsoft interview reports.
“Given a list of words and a string, output the list of words that can be formed from the string’s characters while keeping the character order.”
Read reports →“Three friends in Seattle say it's raining. Each one lies with probability 1/3. What is the probability that Seattle is actually rainy?”
Read reports →“There are 6 marbles in a bag and 1 of them is white. You reach into the bag 100 times, putting the marble back each time. What is the probability that you draw the white marble at least once?”
Read reports →“In a 15-minute time window, the chance of seeing a shooting star or a bunch of them is 0.2. What is the percentage probability of seeing at least one shooting star if you stay under the sky for about an hour? Suppose you have a medical test for a rare disease that is 99% accurate, and the disease occurs in 1% of the population. What is the chance that a person who tests positive really has the disease?”
Read reports →“You’re about to fly to Seattle and want to know if you should bring an umbrella. You call 3 random friends who live there and each independently tells you if it’s raining. Each friend has a 2/3 chance of telling the truth and a 1/3 chance of lying. All 3 say “Yes” it’s raining. What’s the probability that it’s actually raining in Seattle?”
Read report →“How does a random forest model function? What is gradient boosting (for example, XGBoost or LightGBM)? What is the difference between bagging and boosting?”
Read report →“What is backpropagation in neural networks? What are activation functions and why do they matter? How are CNNs and RNNs different? What is dropout and what is the reason for using it?”
Read report →“Explain the 5 key (The 5 Vs) factors to consider when building Big Data pipelines: Volume, Velocity, Variety, Veracity (data quality), and Value. Describe techniques to check for Veracity.”
Read report →“Explain right-skew and left-skew distributions (draw on whiteboard), and where the mean and median values are in these cases compared to a symmetrical normal bell shape.”
Read report →Formats, difficulty and experience
Across all 374 Microsoft interview reports.