Meta Interview Questions — Page 233

Meta·Machine Learning Engineer·Technical Phone Screen

Jul 2026

Meta ML Engineer interview with a coding question that looked straightforward but had enough edge case surface area to trip you up if you weren't careful. Felt like a technical phone screen vibe, focused purely on the algorithm.

Given an m x n integer matrix, check whether every top-left to bottom-right diagonal contains identical values. Solve it in O(mn) time, discuss space trade-offs, and handle edge cases like empty matrices, single row or column, and negative values.

“I jumped straight to the main logic and forgot to ask about negatives until they brought it up themselves, which was a bit embarrassing.”

View Post

Meta·Software Engineer·Technical Phone Screen

Jul 2026

Meta data engineer technical screen, one SQL question the whole time. Pretty focused on cumulative metrics logic which I wasn't expecting to be the entire interview.

Given a daily_metrics table and a cumulative_metrics table that only holds yesterday's data, write SQL to compute today's cumulative value per content_id for a target date D, covering content_ids that appear in either table.

“The FULL OUTER JOIN part is what tripped me up at first.”

View Post

Meta·Software Engineer·Onsite - System Design / Architecture

Jul 2026

Meta system design screen for a software engineer role, focused entirely on a frequency tracking problem with two distinct workload scenarios. The question had a lot of depth to it and the conversation went pretty long.

Design a data structure that supports adding item occurrences and querying the top-K most frequent items. Propose different designs optimized for read-heavy vs write-heavy workloads, analyze time and space complexity for each, and explain the trade-offs.

“This one looked straightforward at first.”

View Post

Meta·Software Engineer·Onsite - Coding / Algorithms

Jul 2026

Meta SWE coding round, one problem the whole session: kth largest element with a follow-up on handling data that doesn't fit in memory. Pretty standard algorithmic interview but the follow-up is where things got interesting.

Given an integer array that may contain duplicates, find the k-th largest element by value. Implement an average O(n) solution, then discuss or implement an alternative approach for when the dataset is too large to fit in memory.

“The quickselect part I had down.”

View Post

Meta·Data Scientist·Onsite - Product Sense / Strategy

Jul 2026

This was a Meta DS onsite focused entirely on a single sprawling product case about Instagram Shopping. Three parts, all connected, and they expected you to hold the whole thing together across hypothesis validation, revenue sizing, and post-launch diagnosis. Brutal in scope but kind of interesting if you like this stuff.

Instagram is considering a new Shopping tab but can't directly observe off-platform purchases. What in-app signals would you use as proxies for off-app purchase intent, and how would you account for bias and build counterfactuals to validate the hypothesis that users discover products on Instagram but buy elsewhere?
Using the given assumptions (50M DAU, 10% weekly intent, 0.2 purchases per intentor at $45 AOV, 5% take rate, plus ad impression fees), estimate the weekly gross revenue from transaction fees vs. ads, then identify which three assumptions most dominate the estimate and describe what sensitivity analyses you'd run.
After launching the Shopping tab, you see high click-through into merchant flows but purchases are way below forecast. Build a metric tree and investigation plan that localizes funnel drop-offs, evaluates cannibalization vs. incrementality, separates UX friction from supply/quality issues, and rules out data instrumentation problems. Also specify what would trigger a rollback.

“This part tripped me up more than I expected.”

View Post

Meta·Data Scientist·Technical Phone Screen

Jul 2026

Meta DS interview focused almost entirely on experiment design for a social feature, specifically a friend-recommendation algorithm. The depth they expected was genuinely surprising, way beyond 'pick a metric and run a t-test'.

You're evaluating a new friend-recommendation algorithm behind a feature flag. What are your primary success metrics and guardrail metrics, and how do you define attribution windows for each?
How would you define the experimental unit and handle network interference effects in this experiment?
Given these 14-day results: MAU up 5% (significant), engagement per user down 3% (significant), confirmed friends per DAU up 2% (not significant at p=0.07), would you launch or not? Walk through your decision rule.
Translate a 2% lift in confirmed-friends-per-DAU into an estimated annual revenue impact. What's the minimum effect size that would justify a full rollout?
After launch, how would you monitor for regression and what would trigger an automatic rollback?

“I went with confirmed-friends-per-DAU, acceptance rate, and qualified exposures-per-DAU as primaries, with a 14-day window.”

View Post

Meta·Data Scientist·Technical Phone Screen

Jul 2026

A Meta DS interview that went deep on experiment design for a bot-mitigation system. The question was basically one giant case study covering randomization, metrics, power, and fallbacks, and it kept branching in ways I wasn't fully ready for.

You're rolling out a system that hides or rate-limits comments from suspected bots. How would you design an experiment to measure its impact, and how do you justify your randomization unit given potential spillovers?
What are your primary success metrics and guardrail metrics for this bot-mitigation experiment?
How would you handle already-flagged accounts and high-risk geographies in your experiment eligibility criteria, and what variance reduction techniques would you apply?
Walk through your power and sample size calculations for this experiment, given a baseline of 5.5 human-visible comments per DAU, a 14-day test window, and day-level clustering.
How would you account for novelty effects, bot migration to the control group, and adversarial adaptation over the course of the experiment?
If a proper A/B test isn't feasible, what alternative causal inference approach would you use, and how would you validate its assumptions?
If false positive rates rise during the experiment, how would you diagnose whether the bot-mitigation system is incorrectly flagging real users, and what segmentation would you apply?

“This is where I got tripped up first.”

View Post

Meta·Software Engineer·Technical Phone Screen

Jul 2026

Meta SWE coding round with a median-from-stream problem. Pretty standard algorithmic interview but the real-time constraint pushed it beyond a naive sorted array approach.

Design a data structure that supports adding numbers from a stream one at a time and returning the current median after each insertion. How do you keep this efficient as the stream grows?

“My first instinct was binary search insertion into a sorted array which is what the prompt basically hints at, but then they asked about the time complexity and I had to admit that shifting elements on insert is O(n).”

View Post

Meta·Software Engineer·Technical Phone Screen

Jul 2026

Meta SWE coding round, one sliding window problem about maximizing vacation days by converting workdays to holidays with a fixed PTO budget. Pretty clean problem once you see it, but I second-guessed myself on the edge cases more than I should have.

Given a string of 'H' (holiday) and 'W' (workday) characters and an integer PTO, find the length of the longest contiguous subarray where the number of 'W' days is at most PTO.

“Classic sliding window once you reframe it: you're just looking for the longest window with at most PTO occurrences of 'W'.”

View Post

Meta·Data Scientist·Technical Phone Screen

Jul 2026

Meta data scientist interview with a meaty stats/optimization question about threshold selection on a skewed score distribution. One question, lots of sub-parts, felt more like a take-home problem crammed into a live session.

You have a per-user interest score between 0 and 1 for a new feature, and the distribution is right-skewed. You can only contact at most K users per week. How would you pick a principled score threshold to maximize incremental value, given a per-contact cost and an expected benefit function? Walk through how you'd estimate the benefit function from historical data, find the profit-optimal cutoff, handle a left-skewed distribution, deal with sampling noise, and set up an ongoing backtest.

“This one has five sub-parts and I fumbled the ordering badly.”

View Post