Jobra
🐞Step 1 of Jobra:AI Debugging Rounds
DSA+AIPreparing you for both.
JOBRA · AI DEBUG ROUND (PREVIEW)sum.cpp
// buggy snippet
int total(vector<int>& a, int n) {
    int sum = 0;
    for (int i = 0; i <= n; i++) { // 🐞        sum += a[i];
    }
    return sum;
}
A real codebase. A real bug. You + AI.

Jobra drops you into a broken repo. Find it, fix it, ship it. AI helps you debug like on the job.

Prompt:Correct the loop condition
Tap 'Let Jobra fix it' below to simulate AI debugging...

How to Crack Meta: The Software Engineer Interview, From Real Experiences

7 min read

Prepare for Meta software engineer interviews with reported assessment patterns, interview questions, and preparation advice.

In this article
On this page

Last reviewed November 15, 2026.

For new grads, interns and experienced software engineers.

Built by the Codinzhub team from Meta interview experiences that candidates shared. Each experience is one person's interview, and Meta changes its process by year, level and team, so use this as a map, not a promise.

The 30-second version. A recruiter call, sometimes a CodeSignal OA, then a 45-minute phone screen with two coding problems, then a virtual onsite of four or five rounds: two coding rounds (two problems each), system design or product architecture, and behavioral. Since 2026 many loops replace a coding round with an AI-enabled coding round and add an AI-enabled design round. Then team matching. Meta values speed: about 90% of problems are mediums, and two in 45 minutes is tight.

Round-by-round breakdown

Stage 1: Application, where candidates applied

How it works. Most candidates were contacted by a recruiter or had a referral (from an alum or a former colleague). Others applied on LinkedIn or the careers site. One candidate sent about 100 applications with little response, then got in through a referral. New grad recruiters said some roles were urgent and had to onboard within months.

Stage 2: Shortlisting, how candidates moved forward

How it works. A recruiter call about your background, level and team preference. Several candidates were offered a mock interview first, which does not count. You can ask for preparation time: candidates scheduled the screen about a month out. One candidate's timeline from referral to signed offer was 205 days.

Stage 3: Online assessment (some candidates)

At a glance

  • Length: about 90 minutes on CodeSignal, camera on
  • Questions per round: 4 levels of one problem, or 4 separate problems
  • What was asked: class design that grows level by level, simulations, arrays
  • Most common: the bank system; the in-memory database

How it works. A level-up format: you only see the next level after passing the current one's tests, and each level can break your earlier data structures. One new grad assessment also had 40 questions, 35 of them about how you use AI.

Questions candidates reported:

Question (our summary)What it tests
Bank system: accounts, top N spenders, scheduled transfers, merge accountsClass design, heaps
In-memory database: set and get, compare-and-set, scan by prefix, TTL, read at an earlier timeHash maps, versioning
Tetris-style placement in a grid; longest run of houses after each buildSimulation, union-find
Best rating-to-price ratio; round trips between two schedulesArrays, two pointers

More questions: Browse Meta questions.

Stage 4: Phone screen

At a glance

  • Length: 45 minutes (about 5 minutes of introduction)
  • Questions per round: 2 coding problems, often with follow-ups
  • What was asked: grids and BFS, prefix sums, trees, heaps, strings
  • Most common: Kth largest element; lowest common ancestor; subarray sum equals K

How it works. Two timed coding-problem problems, often Meta-tagged and slightly changed. You cannot run the code, so write test cases and dry-run them. One candidate solved both optimally and was still rejected because they did not explain their thinking out loud.

Questions candidates reported:

Question (our summary)Follow-up reported
Kth largest element; lowest common ancestorWhat if it is a BST; what if nodes have parent pointers
Shortest path in a binary gridBFS vs DFS
Count subarrays that sum to a targetPrefix sums with a hash map
Merge three sorted arrays without duplicates; simplify a cd pathApply it to the current directory
Nearest equal values within distance kSave space

More questions: Browse Meta questions.

Stage 5: The virtual onsite

How it works. Four or five 45-minute rounds, often over two days. Coding and design rounds have no self-introduction. New grads had three coding rounds and one behavioral round.

Coding rounds

At a glance

  • Length: 45 minutes
  • Questions per round: 2 problems, one usually with a harder follow-up
  • What was asked: binary search, trees, graphs, strings, design-a-container problems
  • Most common: valid palindrome with deletions; local minimum by binary search; tree traversals

Questions candidates reported:

Question (our summary)Follow-up reported
Can the string be a palindrome after deleting one character?Delete up to k characters
Find a local minimum (Find Peak Element variant)Iterative only; fewer comparisons
Is there an island of exactly this size?Precompute island sizes
Course schedule where each course takes hours: minimum hours for given targetsDisconnected graphs
Insert into a sorted circular linked listStop when you return to the start
Sum of BST values in a rangeAverage instead; many queries
Sparse vector dot productOne vector dense; no hash maps
Container with insert and remove-random in O(1)Swap with last, then pop

More questions: Browse Meta questions.

AI-enabled coding round

At a glance

  • Length: about 45 to 60 minutes in CoderPad, with an AI assistant
  • Questions per round: 1 problem in 3 steps: fix a bug, implement, optimise
  • What was asked: a card game, word subsets, friend recommendations, a maze with keys and doors
  • Most common: pick three cards that sum to 15; maximum unique-letter word subset

How it works. A small multi-file project with test suites, the largest with about 100,000 users and 1.5 million friendships. The sandbox is slow (about 1.5 cores), so do not let the AI run the large tests. You must show you understand the AI's code.

Questions candidates reported:

Question (our summary)What it tests
Card game: pick three cards summing to 15 from a visible deckDebugging, search, benchmarking
Largest set of words with no shared lettersBitmasks, pruning
Recommend friends from a friendship graphGraphs at scale
Maze solver; then keys and doorsBFS with state

System design or product architecture

At a glance

  • Length: 45 minutes
  • Questions per round: 1 design
  • What was asked: leaderboards, search, feeds, ticketing, ride sharing, agents
  • Most common: a game leaderboard; searching posts

Questions candidates reported:

Question (our summary)What the interviewer focused on
Game leaderboard showing the closest scores after a gameRange queries over a large dataset
Search text status updates with AND and ORInverted index, sharding
Ticketmaster with a 5-minute holdACID and isolation
Design coding-practice with contestsSandboxed judges, scaling
An agent that triages tickets (AI-enabled design)Where an LLM is needed, cost per call

More questions: Browse Meta questions.

Behavioral round

At a glance

  • Length: 45 minutes
  • Questions per round: 5 to 10 questions with follow-ups
  • What was asked: ambiguity, conflict, feedback, proudest project, AI use
  • Most common: handling conflict; working with incomplete information

Questions candidates reported: how you deal with ambiguity; a conflict and how you resolved it; your proudest recent project; critical feedback and what you changed; a project that changed direction; a difficult coworker; AI-driven impact and how you keep learning AI. For new grads the round did not ask about the resume.

More questions: Browse Meta questions.

Stage 6: Team match and offer

How it works. Rejections usually came within four business days. Passing candidates went to team matching, with up to four calls; it took a week for some and months for others. One E4 candidate solved every problem and was still rejected for not meeting E4 expectations. One candidate got the offer despite a weak signal on one coding round.

Topics to prioritise

Across these experiences, the most common topics are trees, graphs and BFS, binary search, strings and palindromes, heaps, hash maps with prefix sums, and system design. Meta rarely asks DP, but new grad follow-ups did.

The Codinzhub Next-Round Playbook

What Meta candidates who moved on did. Put together by the Codinzhub team from the Meta experiences in this guide.

To get past the phone screen

  • Practise the Meta-tagged list by frequency until two mediums fit in 40 minutes.
  • Build up from one approach instead of listing three at once; it looks memorised.
  • Write your own tests: empty, one element, two, all equal, increasing, decreasing, duplicates.

To get past the coding rounds

  • Say when you are stuck. A hint early is better than five silent minutes.
  • Leave time for the follow-up; at new grad level it is often the hard part.

To get past the AI-enabled round

  • Practise in a slow sandbox, and tell the AI to work in small steps and skip large test runs.
  • Explain each change the AI makes.

To get past behavioral

  • Prepare six or seven stories and expect deep follow-ups on what exactly you did.

FAQ

Is DP asked at Meta?

Rarely, but a new grad candidate was asked to code a delete-k-characters palindrome follow-up with DP.

Can I run code in the coding rounds?

No. Candidates dry-run by hand. The AI-enabled round is the exception.

About this guide

This guide is researched and written by the Codinzhub team. We read every Meta interview experience we could find, kept only what candidates actually described, and rewrote it in our own words. Questions are our own short summaries; the practice links point to similar problems, not the exact interview questions. Found something out of date? Share your own experience on Codinzhub and we will update the guide.

Ready for the next topic?

Keep track of what you’ve learned.