Minimum Remove to Make Valid Parentheses (LC 1249): given a string of '(' , ')' and letters, remove the minimum number of parentheses to make it valid and return any valid result.

Clarify that letters are untouched and any valid output is fine. Single left-to-right pass with a stack/index set: track unmatched '(' indices, mark stray ')' for removal, then remove leftover '(' at the end. Rebuild the string from surviving indices. O(n) time, O(n) space; state it and dry-run an unbalanced example. Whether you reach an O(n) one-pass solution cleanly, handle edge cases (only letters, all closers), and verify on examples rather than guessing. This is one of Meta's single most-reported problems.

Valid Palindrome II (LC 680): given a string, return true if it can become a palindrome by deleting at most one character.

Two pointers from both ends; on the first mismatch, branch into two checks — skip the left char OR skip the right char — and return true if either remaining substring is a palindrome. Factor out a helper isPalindrome(i, j). O(n) time, O(1) extra. Explicitly handle the already-palindrome and empty-string cases. Recognizing the at-most-one-deletion branch instead of brute-forcing all deletions, clean helper decomposition, and correct two-pointer bookkeeping.

Binary Tree Vertical Order Traversal (LC 314): return the nodes of a binary tree grouped by column, top-to-bottom, left-to-right; nodes in the same row and column are ordered as they appear left-to-right.

BFS (not DFS) with a queue of (node, column) so same-row nodes keep left-to-right order; bucket values into a hashmap keyed by column, tracking min/max column. Emit columns in sorted order at the end. Call out why BFS beats DFS here (DFS would need an extra row/depth tiebreak). O(n) time. Choosing BFS for stable ordering, correct column indexing (left = col-1, right = col+1), and handling the same-cell ordering tiebreak. A classic Meta tree question.

Design an Ad Click Aggregator: ingest a high-volume stream of ad-click events and serve near-real-time aggregated metrics (clicks per ad over time windows) on a dashboard.

Sketch the path quickly (clickers → ingestion/Kafka → stream aggregation → OLAP store → query API) then spend the time on the hard parts: reducing click-to-query latency, exactly-once vs at-least-once and idempotency/dedup, windowed aggregation, hot-partition/celebrity-ad skew, and reprocessing for late or corrected data. Justify each technology choice. Depth on streaming aggregation, fault tolerance, idempotency, and scaling — Meta wants you to blow through the high-level design and go deep on trade-offs. One of the most-reported Meta infra prompts.

Design a Top-K / trending-topics service: return the top K most frequent items (hashtags, search terms, songs) over a sliding time window at very high throughput.

Distinguish exact vs approximate; propose Count-Min Sketch + per-shard heaps merged at query time for the approximate path, partitioning by key, and time-window bucketing. Discuss accuracy-vs-cost trade-offs, hot keys, and how K and the window size change the design. Address read vs write amplification. Comfort with probabilistic data structures, sharding/merge strategy, and articulating the approximation trade-off rather than insisting on exactness.

Design Facebook Live Comments: broadcast comments on a live video to a huge fan-out of concurrent viewers in real time.

Decide the transport (WebSocket/SSE vs long-poll), a pub/sub fan-out layer, and how to shard connections per live video. Handle backpressure and rate-limiting on viral streams, ordering vs eventual delivery, and how late joiners get recent history (e.g. a ring buffer of recent comments snapshotted to Redis/CDN). Estimate connection counts and discuss horizontal scaling of the gateway tier. Real-time fan-out reasoning, connection/state management at scale, and graceful degradation under spikes — a Meta-specific scenario tied to their product surface.

Tell me about a time you had a conflict with a coworker. How did you resolve it?

Use a tight STAR story with a real technical disagreement. Show you sought to understand their view (empathy), brought data, and drove to a resolution that benefited the company/team — not just 'I won.' Close with the outcome and the working relationship afterward. Meta's culture prizes direct, candid feedback, so don't present yourself as conflict-avoidant. Conflict resolution + empathy: a positive, company-benefiting outcome, willingness to engage directly, and reflection. Interviewers probe for a resolution that helped the org, not a personal victory.

Tell me about the project you're most proud of and why.

Pick a project with measurable impact at team scale (3+ people for E5). Lead with the impact, then your specific contribution, the hardest technical/ambiguity challenges, and what you'd do differently. Quantify (latency, revenue, users) and make YOUR role unambiguous within the team effort. Motivation and scope of ownership/impact — and clear, structured storytelling that separates 'I' from 'we'. For E5 they calibrate level off the scope you can credibly own.

Tell me about a project or set of requirements that was ambiguous or underspecified, and how you adapted as it kept changing.

Show how you made progress without full clarity: framed assumptions, broke the problem down, aligned stakeholders, and shipped iteratively while requirements shifted. Emphasize the bias-to-action and how you de-risked rather than waiting for perfect specs. Working in an unstructured environment + proactivity + perseverance: comfort with ambiguity, driving forward, and building consensus — explicitly weighted heavily for E5+.

Meta Interview Questions

Meta's SWE loop (2025-26) is: recruiter call → CodeSignal online assessment (90 min, video/mic-monitored) → technical phone screen (2 medium problems in ~35-45 min) → onsite of 4 rounds (E5): one traditional coding, one new AI-enabled coding round (piloted from Oct 2025), one design round, and one behavioral ("Jedi") round. The design round is System Design for Infra roles and Product Architecture for Product roles. Meta's coding bar skews heavily to LeetCode mediums (rarely DP), with ~2 questions per 45-min round.

28 real questions across 3 rounds

Reported via Reported on the LeetCode Meta company tag, HelloInterview Meta E5 guide (lists LC 1249/314/227/215 as most-asked), DesignGurus 'Top Meta', FinalRound 'leaked Meta questions', and the widely-cited Medium 'Meta's 73 most-asked' list. Meta is ~90% medium, almost never DP, ~2 questions per 45-min round. · Meta's AI-enabled coding round, piloted from Oct 2025 (HelloInterview 'Meta's AI-Enabled Coding' blog, Coditioning, interviewing.io, CoderPad). ~60 min in a 3-panel CoderPad with a built-in AI assistant (model picker incl. GPT-5/Claude Sonnet 4.5/Gemini 2.5 Pro/Llama 4); AI has file context but cannot edit your code. Replaces one of the two traditional coding rounds.

10 more coding questions Meta actually asks, each with how to approach it and exactly what the interviewer is evaluating. Unlock the full set across every round.

Continue with Google

Or run a Meta mock with the AI interviewer. It asks questions like these, follows up on your answers, and tells you exactly what to fix. The only score that means anything is the one on your real answers.

Continue with Google

Keep prepping

All company interviews