Meta Interview Questions — Page 46

Meta·Software Engineer·Onsite - System Design / Architecture

Apr 2026

Meta system design slash behavioral hybrid, at least that's how it felt. The one question they asked was a beast and I don't think I fully recovered from the second half of it.

Walk me through a recent project where you had significant impact. What was the problem, what options did you consider, what trade-offs did you make, what was your specific role and technical contribution, how did you influence stakeholders outside your team, and what measurable results came out of it?

“This is like five questions stapled together and they want a coherent answer.”

View Post

Meta·Data Scientist·Technical Phone Screen

Apr 2026

Meta DS interview focused entirely on one deep system design question about replacing a legacy ads ranking model. The whole thing was a multi-part beast covering offline eval, online experimentation, and post-launch monitoring. Felt more like a product/ML systems design loop than a typical DS screen.

You have a legacy ads ranking model and a new challenger model built by another team. Walk through your complete plan to validate and replace the old model, from offline checks all the way to production rollout.
How would you use offline logs from the old model to estimate how the new model would have performed, and what are the assumptions and limitations of that approach?
Your experiment shows CTR dropped about 1.5% but revenue per thousand impressions went up 3% and advertiser return on ad spend improved. How do you decide whether to ship the new model?
What does your ramp schedule look like for rolling out the new model, and what would trigger a rollback at each stage?
What are the main production risks after the new model is fully launched, and how would you monitor for them?

“This question is massive and I underestimated how structured they wanted the answer.”

View Post

Meta·Product Manager·Onsite - Product Sense / Strategy

Apr 2026

Meta PM interview for Horizon Worlds, product sense round focused on metrics. One question, no fluff, straight into it.

You're the PM for Horizon Worlds. Which metrics do you prioritize first?

“I went straight to DAU and time-in-world, which felt safe but probably too surface-level for a product that's still trying to prove out its core loop.”

View Post

Meta·Software Engineer·Technical Phone Screen

Apr 2026

Meta SWE screen, one coding problem, pretty standard fare for that company. The question itself was clean but the edge cases are where things get interesting.

Given a string, return true if it can be made into a palindrome by removing at most one character. Expected O(n) time.

“Two pointers from both ends, compare characters, and when you hit a mismatch you get one free skip.”

View Post

Meta·Software Engineer·Technical Phone Screen

Apr 2026

Meta SWE phone screen that basically turned into a grid DP session with a sneaky robot-maze follow-up hiding in the back pocket. Two variants of the same core problem, and the gap between them is bigger than it looks on paper.

Given an m x n grid where each cell holds a reward value (which can be zero or negative), find the maximum total reward you can collect along a path from the top-left to the bottom-right corner, moving only right or down.
Same maze/reward problem, but now you only have move() and canMove() API calls and a isCheese() check. No global grid view. How do you find the goal and maximize reward collected?
Follow-up: what if the robot can now move in all four directions instead of just right and down? Does your DP approach still work?

“Pretty standard once you see it.”

View Post

Meta·Software Engineer·Technical Phone Screen

Apr 2026

Meta SWE interview with a tree traversal problem that had a twist baked in. Pretty focused session, one meaty coding question with follow-ups on complexity and design choices.

Given a binary tree and a predicate function P that evaluates node values, return a list of lists where each inner list holds the values at a given depth that satisfy P. You must implement this iteratively using a stack, no recursion allowed. Explain how you track level boundaries, manage per-level results, and give a complexity analysis.

“The predicate filter part was fine, the annoying bit was detecting level boundaries without recursion.”

View Post

Meta·Software Engineer·Onsite - Behavioral / Leadership

Apr 2026

Behavioral round at Meta for a software engineer role. Pretty standard set of questions, nothing that surprised me, but I definitely fumbled a couple of them more than I expected to.

Walk me through your background and what's been driving your career decisions.
Tell me about a project that was particularly difficult and how you got through it.
Describe a time you disagreed with a teammate. How did you handle it?
Can you share an example of receiving critical feedback and what you actually did with it?
Why this role, and why Meta specifically?

“I over-rehearsed this one and it showed.”

View Post

Meta·Data Analyst·Technical Phone Screen

Apr 2026

Interviewed at Meta for what seems like a data or analytics role. The content is pretty sparse so there's not much to go on, but the question about sales by customer city points to some kind of SQL or product analytics exercise.

Write a query to show total sales broken down by customer city.

“Pretty standard aggregation question.”

View Post

Meta·Software Engineer·Onsite - System Design / Architecture

Apr 2026

System design round at Meta for a software engineer role. The whole session was one big deep-dive into a customer merge operation for a payments platform, and they really did expect you to go nine levels deep on it.

Design a mergeCustomers(oldId, newId) operation for a payments and ledger platform. The merge absorbs the old identity into the new one while keeping every account's balance and full transaction history intact, with accounts staying attached to their own IDs rather than being merged together.
How do you unify two customer identities so that both the old and new IDs resolve to the same canonical identity for all reads and future writes, starting from the moment the merge commits?
How do you reassign ownership metadata across all entities that reference a customer (accounts, payment instruments, transfers, beneficiaries, etc.) without rewriting the ledger itself?
How do you re-link historical payments and transfers, plus any recurring or scheduled items, so nothing gets lost or executed twice after the merge?
How do pending or held transactions, authorizations, and in-flight transfers continue to settle correctly after the merge completes?
How do you handle conflicts and duplicates that arise from the merge, like duplicate payment methods on the same token, identical scheduled items, beneficiary collisions, single-primary constraints, and per-customer unique nickname conflicts? What are your deterministic resolution policies?
How do you update or invalidate indexes and caches after a merge, including top-N payers and top-N spenders leaderboards, so aggregates stay correct immediately?
How do you ensure that balances, transaction history, and top-N aggregates remain correct with no loss and no double-counting for queries issued using either the old or new customer ID after the merge?
What atomicity and consistency guarantees does your merge operation provide? How do you make it idempotent so repeating the same merge is a no-op, and how do you handle concurrent merges including chain scenarios like A merging into B and then B merging into C?
What is the time and space complexity of the merge operation itself and of the post-merge read path?

“This was the whole interview, basically.”

View Post

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

Apr 2026

ML system design round at Meta for an MLE role. One meaty question about copyright detection that branched into a bunch of edge cases I wasn't fully ready for.

Design a machine learning system to detect whether a user-uploaded image infringes on copyright, including edge cases like collages where only a portion is infringing, images with adversarial text overlays meant to evade detection, and near-duplicate photos of copyrighted works.

“The base case felt manageable: perceptual hashing, embedding similarity, maybe a classifier on top.”

View Post