phone screen, followed by 4 back to back technical and coding, interviewers seems to be very tired and not quite energetic. questions were not related to DS position rather software engineering, networks, search optimization,...... To me it seems they just were trying to get it done, except 1 interviewer the rest were quite amateur which was weird how they were assigned for interview
J'ai postulé en ligne. Le processus a pris 5 semaines. J'ai passé un entretien chez Microsoft (Seattle, WA) en févr. 2026
Entretien
1 hr screening, BQ and logistic questions
1 tech screening, python, sql, probability
3 back to back loop interviews. Mine loop was online, but know some other candidates was asked to be interviewed on-site (Seattle).
En el proceso de entrevista de Microsoft. Pasé la primera ronda telefónica pero luego no me llamaron a continuar para ninguna ronda sucesiva. A pesar del buen feeling del principio, no me llamaron para continuar.
J'ai postulé via la recommandation d'un employé. J'ai passé un entretien chez Microsoft en déc. 2024
Entretien
I applied internally for this position recently and I got a short meeting with the hiring manager whether if I would be a good fit or not. This position was for product design, and they were looking for someone with prior experience in terms of analytics, KPI design, agent-based modeling simulation and specifically feature importance analysis. After our short meeting and the hiring manager set up for meetings one with product manager, head of software engineering, the head of data science and machine learning team and the last one with himself again. It was a nice and eye opening interview but I realized that they were looking for someone with very strong fundamentals with the level of depth that you could teach everything from scratch and you should be able to explain everything with pros and cons with prior experience with it, hence it was a good epiphany for me to change my approach about how to prepare for an interview.
Questions d'entretien [3]
Question 1
Product:
Imagine there’s an e-commerce website like Amazon that was launched two years ago, and they have been selling products meanwhile collecting reviews from those who have purchased the products in terms of the 1-5 stars and the text reviews. They have decided after two years to put the reviews that we have been collecting on the website and evaluate if they will improve the sales.
They designed an A/B testing in which there are 50% control group and 50% treatment group. After they ran the test, they realized that the sales had dropped. What do you think caused the sales drop even though the reviews have been provided to the treatment group?
Notes:
• There is no novelty effect.
• Both groups are completely randomized.
• User experience has been the same meaning, there is no extra effort to place an order.
• Those in the treatment group who have purchased products are more statistically satisfied than those in the control group.
• Users have sorting and filtering options over the reviews.
Stats:
1. Fundamental laws
1.1. Explain Central Limit Theorem (CLT)?
1.2. Explain Law of Large Number (LLN)?
1.3. What are their differences? How are they beneficial?
2. Statistical Tests
2.1. Tell me the differences/conditions between T-Test vs Z-Test are? When is each of them used?
2.2. When is t-distribution used as opposed to normal distribution?
2.3. How many data points are considered good enough to use each of them?
2.4. How does each distribution look like? (skewness and kurtosis viewpoint)
2.5. Explain p-value in a layman language with a simple example.
2.6. If we run the t-test multiple times, what will happen to the strength of the statistical test? (Bonferroni Correction)
2.7. When is the Chi-Squared test used? How does the distribution look like?
ML:
1. Linear Regression:
1.1. Explain L1 vs L2?
1.2. How does each affect the coefficients?
1.3. Explain assumptions of linear regression.
1.4. How is each assumption tested?
1.5. If each assumption is violated, what are their remedies?
2. PCA
2.1. Explain PCA.
2.2. Walk me through the algorithm step by step.
2.3. How is the formula constructed?
2.4. What is the relationship between PC1 and PC2?
2.5. How is orthogonality preserved in the mapped feature space?
2.6. How do you run the feature importance in PC-mapped feature space?
3. ML Algorithm
3.1. Explain the ensembling method.
3.2. Explain the differences between XGBoost and Random Forest?
3.3. When is each used? Pros and cons?
3.4. Which one is computationally expensive and why?
3.5. What are the feature selection methodologies?
3.6. Imagine we have a multivariate KPI that most of the features are correlated. Now we are noticing a spike in the KPI, how do you determine which feature has the highest effect on it? (Feature importance analysis for Temporal shock)
