Meta Interview Questions — Page 220

Meta·Data Scientist·Onsite - Product Sense / Strategy

Jul 2026

Meta DS interview centered entirely on a Group Call feature case study, which sounds manageable until you realize it covers like six different dimensions from experiment design to post-launch diagnostics. Dense session, lots of follow-ups.

Using only an existing usage table, how would you decide whether the product actually needs a Group Call feature?
If you had more resources, what additional data or research would you ask for, and why?
How would you set a cap on the number of participants allowed in a group call? Walk through how you'd define and justify that threshold.
Design a full A/B test for the Group Call feature, including hypothesis, metrics, sample size and runtime calculations, guardrails, and potential pitfalls.
Nine months after the Group Call feature launches, what metrics and analyses would you use to evaluate its success?
If the Group Call feature shows no measurable impact on overall company metrics nine months in, is that a problem? Should you keep the feature?
What are the trade-offs between optimizing for ecosystem-level metrics versus focusing on improving the experience for users who already have bad call quality?
Where do you expect users to drop off in the Group Call funnel after launch, and what would you do about it?

“I leaned on proxy signals like conversation size, how often people were initiating sequential one-on-one calls in short windows, that kind of thing.”

View Post

Meta·Data Scientist·Technical Phone Screen

Jul 2026

A Meta data scientist interview focused on product evaluation for a marketplace notification feature. The question was product-heavy with an experimentation angle, which felt more PM-adjacent than I expected for a DS role.

You're evaluating a proposed 'similar listings you may like' notification feature on a marketplace. How would you decide if it's worth building, and how would you define and measure success?

“This is basically two questions stitched together and I didn't realize it until I was halfway through.”

View Post

Meta·Software Engineer·Onsite - Coding / Algorithms

Jul 2026

Meta SWE coding round with two problems back to back. Both were algorithmic, one on BSTs and one on interval merging. Nothing behavioral, just code and complexity analysis the whole time.

Given a binary search tree and an inclusive range [L, R], implement a function that returns both the sum and the average of all node values that fall within that range. You should use the BST property to prune unnecessary traversal. Also handle the edge case where no nodes fall in range, and discuss time and space complexity.
You have two sorted lists of intervals A and B (sorted by start time). Merge them into a single list of non-overlapping intervals that covers the union of all intervals from both lists. Clarify whether touching intervals like [1,3] and [3,5] should be merged, handle overlaps both within and across lists, and analyze the complexity of your solution.

“The pruning part is what makes this interesting and also what trips people up.”

View Post

Meta·Software Engineer·Onsite - System Design / Architecture

Jul 2026

Got a system design question for a Meta EM loop that had nothing to do with Meta's own products, which threw me a bit. The prompt was about building fraud detection for Stripe, so you're basically designing infrastructure for a payments company while sitting in a Meta interview.

Design a fraud detection system for a payments platform like Stripe.

“I went in expecting something Meta-flavored and completely underestimated how deep the payments domain knowledge would matter here.”

View Post

Meta·Software Engineer·Onsite - Coding / Algorithms

Jul 2026

Meta software engineering interview with four coding tasks back to back. The problems ranged from string parsing to tree traversal to OOP design, which is a pretty wide spread for a single session. Felt like a gauntlet more than a conversation.

Given a string, determine whether it represents a valid decimal number, accounting for optional signs, decimal points, and scientific notation exponents.
Given an array of daily asset prices, find the maximum profit from a single buy-sell trade, or return 0 if no profitable trade is possible.
Given a binary tree, return the values of the rightmost node at each depth, as seen from the right side.
Design and implement a card deck library for a standard 52-card deck, including suits, ranks, reset, shuffle, draw, and a remaining count. Handle edge cases like drawing more cards than are left, and write unit tests for it.

“This one is sneakier than it looks.”

View Post

Meta·Software Engineer·Technical Phone Screen

Jul 2026

Meta Production Engineer interview with a coding problem about social graph friend requests. Pretty standard algorithm question but the conditions tripped me up more than I expected.

Given an array of ages for n people on a social media platform, count the total number of friend requests sent, given a specific set of rules about when a person will or won't send a request to another person.

“The three conditions look simple but I kept getting off-by-one errors because of the half-age-plus-seven rule.”

View Post

Meta·Software Engineer·Technical Phone Screen

Jul 2026

Meta Production Engineer interview with a coding problem involving CSV file parsing and physics-based computation. Pretty much a data wrangling plus algorithm question dressed up with dinosaurs, which was a fun twist I did not expect.

You're given two CSV files with dinosaur data. One has name, leg length, and diet. The other has name, stride length, and stance. Using the formula speed = ((stride_length / leg_length) - 1) * sqrt(leg_length * 9.8), compute each dinosaur's speed. Output only dinosaurs present in both files, sorted by speed descending.

“The dinosaur theme threw me off for a second and I spent way too long re-reading the formula instead of just implementing it.”

View Post

Meta·Data Scientist·Technical Phone Screen

Jul 2026

Meta Data Scientist SQL round, one question but it had enough moving parts to keep me busy for a while. The date spine thing tripped me up more than I expected.

Given a table with user composer events (enter, post, cancel), calculate the daily post success rate, defined as posts divided by enters, for each date in a fixed 7-day window ending on 2025-09-01. Every date in the window must appear in the output even with no activity, and the rate should be rounded to 2 decimal places with no integer division.

“The ratio part was fine, NULLIF on the enter count to dodge division by zero, cast one side to float, round to 2.”

View Post

Meta·Software Engineer·Onsite - Coding / Algorithms

Jul 2026

Meta software engineering interview with a heavy algorithms focus. Six coding problems across a single session, ranging from probability and expression parsing to linked list manipulation. No behavioral stuff from what I remember, just back-to-back coding.

Design a class that takes a list of positive integer weights and supports randomly picking an index with probability proportional to each weight. Follow-up: can you do it with O(1) extra space by mutating the input?
Evaluate an arithmetic expression string with +, -, *, and integer division. No parentheses. Target O(1) extra space.
Given a sorted array and a target value, return the first and last index where the target appears. Return [-1, -1] if it's not there. O(log n) expected.
Find the length of the longest valid parentheses substring in a string of '(' and ')'.
Partition a singly linked list around a value x so all nodes less than x come before nodes greater than or equal to x, maintaining relative order within each group.
Count the total number of palindromic substrings in a string.

“The prefix sum approach came to me pretty fast, binary search on top of that for the sampling step.”

View Post

Meta·Data Scientist·Technical Phone Screen

Jul 2026

Meta data scientist interview, looked like a stats/causal inference deep dive. The whole thing was basically one long question about DiD and staggered adoption, which I was not fully expecting to go that deep on.

You can't randomize and need to use staggered adoption across EU regions. Walk through the two-period two-group DiD estimator, then generalize it to multiple periods with staggered treatment timing.
Why does two-way fixed effects produce biased ATT estimates when treatment timing and treatment effects are both heterogeneous? Explain the negative weights problem.
Propose a correct estimator for staggered adoption. Write out the estimand you're targeting and describe how you'd set up an event study with pre-trend tests.
What are the key identification assumptions for DiD, how do you diagnose violations, what standard error approach do you use with few clusters, and how do you handle treatment reversal or partial adoption?
How would you communicate the ATT estimate and its uncertainty to a non-technical executive audience?

“I got the two-group two-period setup fine, that part is pretty standard.”

View Post