Data Scientist • Behavioral • medium

Tell me about a data science project where the results surprised you. What did you do?

#Analytical Thinking

Practice

Data Scientist • Behavioral • medium

Describe how you communicated a complex model result to a non-technical stakeholder.

#Storytelling

Practice

Data Scientist • Behavioral • hard

Tell me about a time you had to push back on a business request for an analysis that would be misleading.

#Ethics #Communication

Practice

Data Scientist • Behavioral • medium

Describe a project where you had to iterate significantly on your initial approach.

#Iteration #Learning

Practice

Data Scientist • Behavioral • medium

How do you prioritize between multiple data science requests from different teams?

#Stakeholder Management

Practice

Data Scientist • Behavioral • hard

Tell me about a time your model failed in production. What did you learn?

#Production #MLOps

Practice

Data Scientist • Behavioral • medium

How do you approach ethical considerations in ML model building?

#Fairness #Bias

Practice

Data Scientist • Behavioral • hard

Describe a time you used data to challenge a widely held assumption in your organization.

#Influence #Analytics

Practice

Data Scientist • Behavioral • medium

Tell me about a time when you strongly disagreed with a Product Manager about the interpretation of an A/B test result. How did you handle it and what was the outcome?

#Conflict Resolution #Stakeholder Management #Communication

Practice

Data Scientist • Behavioral • medium

Tell me about a time when you discovered a significant flaw in your own analysis after you had already presented the findings to leadership. What did you do?

#Integrity #Accountability #Communication

Practice

Data Scientist • Coding • hard

Write a SQL query to calculate 30-day user retention.

#Retention #Analytics

Practice

Data Scientist • Coding • hard

How would you write a funnel analysis query in SQL?

#Funnel #Analytics

Practice

Data Scientist • Coding • medium

Write a query to identify duplicate records and deduplicate them.

#Deduplication #Data Quality

Practice

Data Scientist • Coding • medium

Given two tables: `ad_campaigns` (campaign_id, advertiser_id, spend) and `ad_clicks` (click_id, campaign_id, user_id, timestamp), write a SQL query to find the top 5 advertisers with the highest number of clicks per dollar spent who have spent at least $10,000 in the last 30 days.

#Joins #Aggregations #Date Functions

Practice

Data Scientist • Coding • hard

Given a table `friend_requests` (sender_id, receiver_id, date, status), write a SQL query to calculate the overall friend acceptance rate for each day. Note that a request sent on day 1 might be accepted on day 3.

#Self Joins #Window Functions #Ratio Calculation

Practice

Data Scientist • Coding • medium

Given a pandas DataFrame containing user login logs with columns `user_id` and `login_timestamp`, write a Python function to find the longest consecutive streak of login days for each user.

#Python #Pandas #Data Manipulation

Practice

Data Scientist • Coding • medium

Write a SQL query to calculate the 7-day rolling retention rate of users who signed up through a specific marketing campaign. You have a `users` table (user_id, signup_date, campaign_id) and a `user_activity` table (user_id, activity_date).

#Retention #Rolling Metrics #Joins

Practice

Data Scientist • System Design • hard

How would you build a recommendation system? Compare collaborative vs content-based filtering.

#Collaborative Filtering #Content-Based

Practice

Data Scientist • System Design • hard

Design a real-time fraud detection system for a payments platform.

#Fraud Detection #Real-Time ML

Practice

Data Scientist • System Design • hard

How would you build and deploy a churn prediction model?

#Churn #MLOps

Practice

Data Scientist • System Design • hard

Design a feature store. What are its key components?

#Feature Store #MLOps

Practice

Data Scientist • System Design • hard

We want to test a new feature in Facebook Groups that relies heavily on user interaction. How would you design an A/B test for this feature, keeping in mind that treating one user might affect the experience of another user in the same group?

#Network Effects #Cluster Randomization #A/B Testing

Practice

Data Scientist • System Design • hard

We are evaluating a new machine learning model for the Facebook News Feed ranking. The offline AUC is significantly higher than the production model, but an online A/B test shows no change in user engagement. What could be causing this discrepancy?

#Model Evaluation #Offline vs Online Metrics #Ranking

Practice

Data Scientist • Technical • medium

Explain the bias-variance tradeoff. How does it influence model selection?

#Bias-Variance #Model Selection

Practice

Data Scientist • Technical • medium

What is a p-value? Why is a p-value of 0.05 not always sufficient?

#Hypothesis Testing #p-value

Practice

Data Scientist • Technical • medium

Explain the central limit theorem and its importance in data science.

#CLT #Sampling

Practice

Data Scientist • Technical • easy

What is the difference between Type I and Type II errors?

#Hypothesis Testing #Errors

Practice

Data Scientist • Technical • hard

How do you design an A/B test for a new product feature?

#A/B Testing #Statistics

Practice

Data Scientist • Technical • hard

What is the multiple testing problem? How do you correct for it?

#Bonferroni #FDR

Practice

Data Scientist • Technical • hard

Explain Bayesian vs Frequentist statistics. When would you use each?

#Bayesian #Frequentist

Practice

Data Scientist • Technical • medium

What is a confidence interval? How does it differ from a prediction interval?

#Confidence Interval #Intervals

Practice

Data Scientist • Technical • hard

Explain the curse of dimensionality and its implications for ML models.

#Dimensionality #Feature Engineering

Practice

Data Scientist • Technical • medium

How would you detect and handle multicollinearity in a regression model?

#Multicollinearity #Regression

Practice

Data Scientist • Technical • hard

Explain gradient boosting. How does XGBoost differ from a standard gradient boosting machine?

#Gradient Boosting #XGBoost

Practice

Data Scientist • Technical • medium

How does a Random Forest work? What are its hyperparameters and how do you tune them?

#Random Forest #Hyperparameter Tuning

Practice

Data Scientist • Technical • medium

What is regularization? Explain L1 vs L2 regularization and their effects.

#Regularization #L1 #L2

Practice

Data Scientist • Technical • medium

How do you handle class imbalance in a classification problem?

#Imbalanced Data #SMOTE

Practice

Data Scientist • Technical • medium

Explain the ROC curve and AUC metric. When would you prefer AUC over accuracy?

#ROC #AUC #Metrics

Practice

Data Scientist • Technical • medium

What is cross-validation? Explain k-fold and stratified k-fold.

#Cross Validation #k-Fold

Practice

Data Scientist • Technical • medium

How do you approach feature selection?

#Feature Selection #LASSO

Practice

Data Scientist • Technical • medium

Explain the difference between bagging and boosting.

#Bagging #Boosting

Practice

Data Scientist • Technical • medium

What is principal component analysis (PCA)? What are its limitations?

#PCA #SVD

Practice

Data Scientist • Technical • medium

Explain how backpropagation works.

#Backpropagation #Neural Networks

Practice

Data Scientist • Technical • hard

What is the vanishing gradient problem? How do LSTM and ResNet address it?

#LSTM #ResNet #Gradients

Practice

Data Scientist • Technical • hard

Explain the transformer architecture. What are attention mechanisms?

#Transformers #Attention #BERT

Practice

Data Scientist • Technical • medium

What is transfer learning? How would you fine-tune a pre-trained model?

#Transfer Learning #Fine-Tuning

Practice

Data Scientist • Technical • medium

How would you approach an NLP problem like sentiment analysis from scratch?

#Sentiment Analysis #Text Classification

Practice

Data Scientist • Technical • medium

What is embedding? How do word embeddings like Word2Vec and GloVe work?

#Embeddings #Word2Vec

Practice

Data Scientist • Technical • medium

Explain batch normalization and why it helps training.

#Batch Normalization #Training

Practice

Data Scientist • Technical • medium

How would you detect and mitigate overfitting in a neural network?

#Overfitting #Dropout #Regularization

Practice

Data Scientist • Technical • hard

How would you design an experiment to measure the impact of a new ranking algorithm?

#Experimentation #Metrics

Practice

Data Scientist • Technical • hard

What is a network effect in experimentation? How do you handle SUTVA violation?

#SUTVA #Network Effects

Practice

Data Scientist • Technical • medium

How do you choose a north star metric for a product?

#Metrics #Product Strategy

Practice

Data Scientist • Technical • easy

Explain the difference between a leading indicator and a lagging indicator.

#Metrics #KPIs

Practice

Data Scientist • Technical • hard

How would you identify the root cause of a sudden 20% drop in DAU?

#Root Cause Analysis #Debugging

Practice

Data Scientist • Technical • easy

What is an experiment holdout group?

#Holdout #Control Group

Practice

Data Scientist • Technical • easy

Explain the difference between INNER JOIN, LEFT JOIN, and CROSS JOIN.

#Joins #SQL

Practice

Data Scientist • Technical • hard

How do you monitor model performance in production? What is model drift?

#Model Drift #Monitoring

Practice

Data Scientist • Technical • hard

How does Meta measure the ROI of a Reels algorithm change?

#Reels #Metrics

Practice

Data Scientist • Technical • hard

What is a 'triggered analysis' in the context of Meta's experimentation framework?

#Triggered Analysis

Practice

Data Scientist • Technical • hard

How would you detect bot/spam activity using data analysis at Meta?

#Bot Detection #Analytics

Practice

Data Scientist • Technical • medium

Meta is considering launching a new feature that allows users to tip creators on Reels. How would you determine if this feature is successful?

#Product Strategy #Metrics Definition #Reels

Practice

Data Scientist • Technical • hard

We ran an A/B test on Instagram where we increased the size of the 'Like' button. The results show a 5% increase in likes, but a 2% decrease in comments. Would you launch this change? How do you investigate further?

#Experimentation #Trade-offs #Statistical Significance

Practice

Data Scientist • Technical • medium

You have two coins. One is fair, and the other is biased, coming up heads 75% of the time. You pick one coin at random and flip it three times. It comes up heads all three times. What is the probability that you picked the biased coin?

#Bayes' Theorem #Conditional Probability

Practice

Data Scientist • Technical • medium

The number of daily active users (DAU) on WhatsApp has dropped by 3% week-over-week. Walk me through how you would investigate the root cause of this decline.

#Root Cause Analysis #DAU #Metric Debugging

Practice

Data Scientist • Technical • medium

Facebook Marketplace is seeing a high number of items listed but a low number of transactions completed. What metrics would you look at to understand where the friction is, and what product changes would you propose?

#Funnel Analysis #Marketplace #Conversion Rate

Practice

Data Scientist • Technical • hard

Meta wants to estimate the proportion of fake accounts on the platform. Since we cannot manually review billions of accounts, how would you design a sampling strategy to estimate this proportion with a 95% confidence interval and a 1% margin of error?

#Sampling #Confidence Intervals #Sample Size Calculation

Practice

Data Scientist • Technical • hard

If we introduce a new ad format in the Instagram feed that increases ad revenue by 10% but decreases user time spent by 1%, how do you decide whether to roll this out globally?

#Trade-offs #LTV #Monetization

Practice

Meta

The Interview Loop

Recruiter Screen (30 min)

Technical Loop (3-4 Rounds)

Interview Question Bank

Tell me about a data science project where the results surprised you. What did you do?

Describe how you communicated a complex model result to a non-technical stakeholder.

Tell me about a time you had to push back on a business request for an analysis that would be misleading.

Describe a project where you had to iterate significantly on your initial approach.

How do you prioritize between multiple data science requests from different teams?

Tell me about a time your model failed in production. What did you learn?

How do you approach ethical considerations in ML model building?

Describe a time you used data to challenge a widely held assumption in your organization.

Tell me about a time when you strongly disagreed with a Product Manager about the interpretation of an A/B test result. How did you handle it and what was the outcome?

Tell me about a time when you discovered a significant flaw in your own analysis after you had already presented the findings to leadership. What did you do?

Write a SQL query to calculate 30-day user retention.

How would you write a funnel analysis query in SQL?

Write a query to identify duplicate records and deduplicate them.

Given two tables: `ad_campaigns` (campaign_id, advertiser_id, spend) and `ad_clicks` (click_id, campaign_id, user_id, timestamp), write a SQL query to find the top 5 advertisers with the highest number of clicks per dollar spent who have spent at least $10,000 in the last 30 days.

Given a table `friend_requests` (sender_id, receiver_id, date, status), write a SQL query to calculate the overall friend acceptance rate for each day. Note that a request sent on day 1 might be accepted on day 3.

Given a pandas DataFrame containing user login logs with columns `user_id` and `login_timestamp`, write a Python function to find the longest consecutive streak of login days for each user.

Write a SQL query to calculate the 7-day rolling retention rate of users who signed up through a specific marketing campaign. You have a `users` table (user_id, signup_date, campaign_id) and a `user_activity` table (user_id, activity_date).

How would you build a recommendation system? Compare collaborative vs content-based filtering.

Design a real-time fraud detection system for a payments platform.

How would you build and deploy a churn prediction model?

Design a feature store. What are its key components?

We want to test a new feature in Facebook Groups that relies heavily on user interaction. How would you design an A/B test for this feature, keeping in mind that treating one user might affect the experience of another user in the same group?

We are evaluating a new machine learning model for the Facebook News Feed ranking. The offline AUC is significantly higher than the production model, but an online A/B test shows no change in user engagement. What could be causing this discrepancy?

Explain the bias-variance tradeoff. How does it influence model selection?

What is a p-value? Why is a p-value of 0.05 not always sufficient?

Explain the central limit theorem and its importance in data science.

What is the difference between Type I and Type II errors?

How do you design an A/B test for a new product feature?

What is the multiple testing problem? How do you correct for it?

Explain Bayesian vs Frequentist statistics. When would you use each?

What is a confidence interval? How does it differ from a prediction interval?

Explain the curse of dimensionality and its implications for ML models.

How would you detect and handle multicollinearity in a regression model?

Explain gradient boosting. How does XGBoost differ from a standard gradient boosting machine?

How does a Random Forest work? What are its hyperparameters and how do you tune them?

What is regularization? Explain L1 vs L2 regularization and their effects.

How do you handle class imbalance in a classification problem?

Explain the ROC curve and AUC metric. When would you prefer AUC over accuracy?

What is cross-validation? Explain k-fold and stratified k-fold.

How do you approach feature selection?

Explain the difference between bagging and boosting.

What is principal component analysis (PCA)? What are its limitations?

Explain how backpropagation works.

What is the vanishing gradient problem? How do LSTM and ResNet address it?

Explain the transformer architecture. What are attention mechanisms?

What is transfer learning? How would you fine-tune a pre-trained model?

How would you approach an NLP problem like sentiment analysis from scratch?

What is embedding? How do word embeddings like Word2Vec and GloVe work?

Explain batch normalization and why it helps training.

How would you detect and mitigate overfitting in a neural network?

How would you design an experiment to measure the impact of a new ranking algorithm?

What is a network effect in experimentation? How do you handle SUTVA violation?

How do you choose a north star metric for a product?

Explain the difference between a leading indicator and a lagging indicator.

How would you identify the root cause of a sudden 20% drop in DAU?

What is an experiment holdout group?

Explain the difference between INNER JOIN, LEFT JOIN, and CROSS JOIN.

How do you monitor model performance in production? What is model drift?

How does Meta measure the ROI of a Reels algorithm change?

What is a 'triggered analysis' in the context of Meta's experimentation framework?

How would you detect bot/spam activity using data analysis at Meta?

Meta is considering launching a new feature that allows users to tip creators on Reels. How would you determine if this feature is successful?

We ran an A/B test on Instagram where we increased the size of the 'Like' button. The results show a 5% increase in likes, but a 2% decrease in comments. Would you launch this change? How do you investigate further?

You have two coins. One is fair, and the other is biased, coming up heads 75% of the time. You pick one coin at random and flip it three times. It comes up heads all three times. What is the probability that you picked the biased coin?

The number of daily active users (DAU) on WhatsApp has dropped by 3% week-over-week. Walk me through how you would investigate the root cause of this decline.

Facebook Marketplace is seeing a high number of items listed but a low number of transactions completed. What metrics would you look at to understand where the friction is, and what product changes would you propose?

Meta wants to estimate the proportion of fake accounts on the platform. Since we cannot manually review billions of accounts, how would you design a sampling strategy to estimate this proportion with a 95% confidence interval and a 1% margin of error?

If we introduce a new ad format in the Instagram feed that increases ad revenue by 10% but decreases user time spent by 1%, how do you decide whether to roll this out globally?

Difficulty Radar

Meet Your Interviewers

The "Standard" Interviewer

Unwritten Rules