Meta Interview Questions — Page 3

Meta·Data Scientist·Technical Phone Screen

Oct 2023

Meta DS interview with two SQL questions back to back, both pulling from the same info stream schema. Nothing crazy on the surface but the filtering logic tripped me up more than I expected.

Write a SQL query to find the number of distinct posts that had at least one view longer than 60 seconds from 'Unconnected' viewers within the last 7 days.
Write a SQL query to compute the average number of reactions per post, broken out separately for 'Friend' viewers and 'Unconnected' viewers, over the last 7 days.

“The filtering part is straightforward but I kept second-guessing the date logic.”

View Post

Meta·Data Scientist·Technical Phone Screen

Nov 2023

Meta data scientist interview with a SQL-heavy technical screen. Two questions, both centered on a survey dataset, one about response rates and one about comparing scores across user segments with a significance test thrown in.

Given a survey events table, write SQL to calculate the overall survey response rate.
Write SQL to compare average survey scores between new and existing users, and assess whether any observed difference is statistically significant.

“Straightforward on the surface but I almost forgot to handle the NULLs in survey_score.”

View Post

Meta·Machine Learning Engineer·Onsite - System Design / Architecture

May 2024

Onsite system design round at Meta for an MLE role. Just the one question but it took up the whole session.

Design a video ranking system.

“This is a classic ML system design prompt but the scope creep is real.”

View Post

Meta·Data Scientist·Technical Phone Screen

May 2024

Meta DS interview with a SQL question focused on deduplication logic. Pretty straightforward problem but the business framing around feed quality made it feel more real than a typical leetcode-style thing.

Given a posts table with columns for post ID, user ID, timestamp, and content, write a SQL query that returns the user ID, content, and a count of duplicates for cases where the same user posted identical content on the same calendar day, filtering to only groups with two or more posts.

“The core logic isn't hard once you realize you need to group on three things: user, content, and the date part of the timestamp.”

View Post

Meta·Machine Learning Engineer·Onsite - System Design / Architecture

May 2025

Meta MLE onsite in late May, second system design round. Just the one question but it took the full session and went pretty deep.

Design an event recommendation system.

“Spent probably too long on the retrieval layer and didn't get to re-ranking until the last ten minutes.”

View Post

Meta·Data Scientist·Technical Phone Screen

Aug 2025

Meta DS interview that went deep into statistical testing for a user engagement scenario. Four questions, all quantitative, no fluff. The kind of round where you either know your NB distribution parameterization cold or you don't.

You're comparing daily session counts between new and existing users. The data is integer-valued, skewed, and has lots of zeros. Which test do you use to compare central tendency across the two cohorts and why: two-sample t-test, Welch's t-test, Mann-Whitney U, or a GLM-based approach? Walk through your assumptions and what diagnostics you'd run.
Daily sessions per user follows a Negative Binomial distribution with mean 2.40 and variance 6.96 for existing users. Using the parameterization where E[X] = r(1-p)/p and Var[X] = r(1-p)/p squared, solve for r and p, then compute P(X = 0).
For new users you have mean 1.85 and variance 4.20, with 5,000 users per cohort. Construct a 95% confidence interval for the difference in mean sessions between cohorts. You can use delta-method or GLM reasoning, just state your approximations.
You run a Welch's t-test and get p = 0.04 with Cohen's d = 0.08. How do you interpret statistical versus practical significance? What do you do about multiple testing if you also segmented by 5 countries? And name one robust effect size metric suited to count data, like a ratio of means, and how you'd get a CI for it.

“This is the kind of question where you can talk yourself into a corner fast.”

View Post

Meta·Data Scientist·Technical Phone Screen

Aug 2025

SQL screen for a Data Scientist role at Meta. One question, pretty involved, mixing DAU definitions with a multi-table join across users, events, and calls. Not the hardest thing ever but there are a few gotchas that'll trip you up if you're not careful.

Write a single SQL query to compute the percentage of French DAU (users with any event on 2025-08-31) who participated in at least one video call that same day, either as caller or callee. Return the denominator, numerator, and percentage as a decimal.

“The tricky part is the callee side.”

View Post

Meta·Data Scientist·Onsite - System Design / Architecture

Sep 2025

This was a deep technical screen for a DS role at Meta, centered entirely on one massive open-ended ML system design problem. The question covered basically every angle you can imagine: features, labeling, modeling choices, evaluation, privacy, and deployment. Felt more like a take-home prompt than a 45-minute conversation.

How would you design the features for a model that classifies whether a Facebook session is happening at home, in an office, or at a public venue, using only privacy-preserving signals like timestamps, ASN/ISP type, coarse location, device mix, and contact graph patterns?
How would you generate labels at scale for this network-context classification problem without ground truth, and how do you correct for bias in those noisy weak labels?
Compare rule-based classifiers, gradient boosted trees, and sequence models like HMMs or transformers for this session-context classification task. How do you handle aggregating session-level predictions up to the household level?
How would you evaluate this model? Define your metrics, describe how you'd set up cross-geo temporal cross-validation, and explain how you'd prevent label leakage from Portal adoption data.
What privacy and compliance constraints would you apply to this system, including data minimization, retention limits, on-device inference options, and re-identification risk?
How would you deploy this system, monitor for drift, and design a holdout experiment to measure whether location-type targeting actually improves Portal ad conversion?

“This was the core of the whole interview and it's genuinely hard.”

View Post

Meta·Data Scientist·Technical Phone Screen

Sep 2025

Meta DS interview with a SQL question that looks straightforward until you actually sit down and write it. One question, pretty focused, left me second-guessing my join direction the whole time.

Given a user activity table and a composer events table, calculate the average number of posts per Daily Active User broken down by country for a specific date. DAUs are users with an active flag set on that date, posts are only the 'post' events from the composer table on that same date, and users with zero posts still count in the denominator.

“The zero-post users thing is what gets you.”

View Post

Meta·Data Scientist·Technical Phone Screen

Sep 2025

Meta data scientist interview with a pretty gnarly algorithmic design question about concurrent call statistics. One round, one problem, a lot of moving parts. Walked out not totally sure how I did.

Given N call session intervals [start, end) in UNIX seconds, design an algorithm to compute the 95th percentile of per-minute concurrent call counts over a fixed 24-hour window. The input is a potentially memory-exceeding unsorted stream, duplicate call IDs must be deduplicated by (start, end) pair, half-open interval semantics apply, and the target complexity is O(n log n + M) where M is 1440. Describe your data structures, how you'd handle tie-breaking in the percentile calculation, and how you'd extend this to compute the metric per country if each interval includes a country code.

“This one took me a minute to even parse fully.”

View Post