← Back to Directory

Meta interview notes, page 73 of 238

FAANG / Big Tech

Most of these Meta write-ups come from Software Engineers (1,194), then Data Scientists (489) and Product Managers (453). The most reported rounds are the technical phone screen (1,190), the product sense round (466), the system design round (255) and the behavioral / leadership round (205).

2377 interview notes · updated Jul 2026

Meta·Software Engineer·Onsite - Coding / Algorithms

May 2026
Meta SWE coding round, tree problems the whole way through. The main question was LCA on a binary tree and then they kept piling on variants which I was not fully ready for.
  • Given the root of a binary tree and two nodes p and q (both guaranteed to exist), find their lowest common ancestor. The LCA is the deepest node that is an ancestor of both.
  • If the tree is a BST, how would you use the BST property to find the LCA more efficiently?
  • If each node has a pointer to its parent, how would you compute the LCA more efficiently?
  • If the binary tree is too large to fit in memory, how would you design a system to compute LCA, considering I/O constraints, indexing, blocking, and both preprocessing and online query scenarios?

“Got through this fine with the standard recursive post-order approach.”

View Post

Meta·Software Engineer·Technical Phone Screen

May 2026
Meta Research Scientist technical screen focused almost entirely on the transformer attention mechanism. Dense topic, they went pretty deep and I wasn't totally ready for some of the optimization side of things.
  • Walk through the self-attention mechanism in transformers, including Q/K/V projections, scaled dot-product attention, and why you divide by the square root of d_k.
  • How does multi-head attention work and what's the motivation for splitting into multiple heads rather than using a single large attention operation?
  • How do positional encodings work and why are they necessary in transformer architectures?
  • What are attention masks and how do causal masks differ from padding masks?
  • What is the computational complexity of self-attention and what are the main bottlenecks at scale?
  • How does self-attention differ from cross-attention, and where is cross-attention used?
  • Can you explain FlashAttention and why it's faster than standard attention implementations?

“I started with the matrix projections and felt pretty solid there.”

View Post

Meta·Software Engineer·Technical Phone Screen

May 2026
Meta SWE coding round, got a classic array problem that seems easy until you start second-guessing your indexing. Pretty standard technical screen overall.
  • Given an integer array, implement a data structure that supports repeated range sum queries returning the sum of elements between two indices inclusive.

“Prefix sums, straightforward enough.”

View Post

Meta·Software Engineer·Technical Phone Screen

May 2026
Meta SWE interview with a debugging task on a buggy BFS implementation. Pretty focused and technical, no fluff.
  • You're given a BFS implementation with a bug: the start node is never marked as visited at initialization. Find the bug and fix it, then walk through a test case that demonstrates the fix.

“The codebase was messy enough that I spent way too long just finding where BFS actually kicked off.”

View Post

Meta·Software Engineer·Onsite - Coding / Algorithms

May 2026
Meta SWE coding round with a calculator problem. Pretty standard stuff but the operator precedence angle is where people slip up if they're not careful.
  • Given a string expression containing non-negative integers, '+', '*', and spaces, evaluate it correctly respecting operator precedence (multiplication before addition, no parentheses).

“The multiplication-before-addition precedence is the whole point of the problem and it's easy to just iterate left-to-right and get the wrong answer on something like '3+2*2'.”

View Post

Meta·Software Engineer·Technical Phone Screen

May 2026
Data Engineer interview at Meta with a set of coding problems that mixed practical library/log system design with classic algorithm stuff. Nothing too wild but the log validation one took longer than I expected.
  • Given an array of items each with a category and point value, and an integer k, select exactly k items such that all chosen items belong to different categories and the total points are maximized. What is the maximum sum achievable?
  • Design a data structure to manage book locations in a library, supporting add, move, remove, and get_location operations efficiently. A location can include branch, aisle, shelf, and position.
  • Given a chronological list of library events (timestamp, book ID, member ID, action where action is checkout, return, or renew), validate the log according to these rules: a book can't be held by two members at once, a return must follow a checkout, a renew can only be done by the current holder, and timestamps per book must be non-decreasing. Return a boolean and, if invalid, the index of the first bad event.
  • Determine whether two lowercase strings are anagrams of each other, but without using any counter utility from a standard library. Use only basic data structures like arrays or dictionaries.

“Greedy with a sort felt right to me: sort by points descending, then pick the highest-value item per category until you hit k.”

View Post

Meta·Product Manager·Onsite - Product Sense / Strategy

May 2026
Meta product sense interview, one question about Facebook Dating metrics. Not a lot of context to go on but it's the kind of question that sounds straightforward until you're actually in it.
  • What success metrics would you define for Facebook Dating?

“I went straight for match rate and DAU which felt obvious in retrospect.”

View Post

Meta·Data Scientist·Technical Phone Screen

May 2026
Meta DS interview with a meaty SQL question on content moderation prevalence metrics. One question, but it packed in a lot: window logic, ex-ante vs ex-post definitions, aggregation, and a creator-level breakdown. The kind of problem where you can get the easy parts right and still miss the subtle stuff.
  • Given tables for content views, content moderation decisions, and content metadata, write SQL to compute daily view prevalence of violating content by country over a 7-day window. You need two variants: ex-ante prevalence (a view counts as violating only if a violation decision existed at or before the view timestamp) and ex-post prevalence (based on the final decision regardless of detection timing). Then find the top 3 creators by ex-post violating view share and break down their violating views by surface.

“The ex-ante vs ex-post distinction is where most of the complexity lives.”

View Post

Meta·Technical Program Manager·Hiring Manager Screen

May 2026
Interviewed for a TPM role at Meta, one behavioral question about my last job and that was pretty much it from what I can tell.
  • What did you enjoy most about your last role?

“Seems easy but I fumbled it a bit.”

View Post

Meta·Software Engineer·Onsite - System Design / Architecture

May 2026
Meta system design round for a software engineer role, focused entirely on building a weapon ad detection system end to end. Dense question, lots of ground to cover, and the cold-start angle caught me off guard more than I expected.
  • Design a full end-to-end ML system to detect and prevent weapon sale advertisements, covering ingestion, multimodal feature extraction, labeling, model architecture, online inference, and feedback loops.
  • How would you handle detecting a brand new, never-before-seen weapon category that your training data has never included?

“This is a beast of a question and I underestimated how much they'd want me to go wide before going deep.”

View Post