Max Stack (LeetCode 716): Design a stack that supports push, pop, top, peekMax, and popMax operations, each in O(log n) time in the worst case.

The key insight is that a naive stack gives O(1) push/pop but O(n) popMax; to hit O(log n) for all operations you need a doubly linked list paired with a sorted data structure (TreeMap/sorted set) that maps values to their positions in the list, so you can remove from both structures in O(log n). Walk through the two-structure approach, code all five methods, then discuss the O(n) space tradeoff. A likely follow-up is asking for the O(1) amortized approach using two stacks (one max-tracking stack), and why it fails the O(log n) popMax requirement. Ability to reason about multiple concurrent data structures, clear complexity analysis for each operation, and handling of duplicate values in the sorted structure.

Find Leaves of Binary Tree (LeetCode 366): Given a binary tree, collect and remove all leaves repeatedly until the tree is empty, returning the leaves at each iteration as a list of lists.

The key insight is that the grouping is by a node's height (distance to its farthest leaf child), not by BFS layer. A single post-order DFS computes each node's height and appends its value to result[height], giving O(n) time and O(n) space. A common follow-up asks how output order within each group matters and whether the traversal order can be changed. Recognizing that level-order BFS gives the wrong grouping and that height is computed bottom-up; clean recursive implementation with correct base cases.

Nested List Weight Sum II (LeetCode 364): Given a nested list of integers where each integer's weight is (max_depth - own_depth + 1), return the weighted sum.

The key insight versus the original problem (LeetCode 339) is that the weight is now inverted. Two passes work: first compute max depth with DFS, then a second DFS multiplies each integer by (maxDepth - currentDepth + 1). A single-pass BFS approach tracks an unweighted sum and adds it again for each level, which is more elegant and is the expected follow-up discussion. Both are O(n) time and O(n) space. Understanding the depth-inversion insight, ability to produce both a two-pass DFS and the single-pass BFS accumulation, and discussion of edge cases like flat lists.

Design LinkedIn's News Feed: Design a system that shows a personalized, ranked feed of posts from a user's connections and followed entities for hundreds of millions of users.

Start with requirements: fanout-on-write vs fanout-on-read tradeoffs at LinkedIn's scale (the FollowFeed index handles 80% of feed content). Propose a pull-then-rank pipeline: fetch recent activity from a user's connections via an activity index, rank with a lightweight ML scorer, cache the top-N results per user, and stream updates via a push channel for online users. Key components: activity store (Kafka for events), feed cache (Redis), ranking service, and CDN for media. Follow-ups typically drill into handling celebrities (high fanout), real-time vs eventual consistency, and cache invalidation on post edits. Awareness of the fanout problem and when to choose push vs pull, understanding of feed ranking as a separate concern from retrieval, and a coherent data flow from post creation to user display.

Design a system to surface the top third-party articles shared on LinkedIn (trending content aggregation): Given millions of share events per hour, show the globally trending external URLs in near real-time.

Key insight: exact counting at scale is expensive; use a Count-Min Sketch or a bucketed sliding-window counter (e.g., 5-minute buckets kept for 24 hours) in a stream-processing layer (Kafka + Flink/Spark Streaming). Aggregate per-URL counts into a sorted top-K structure (min-heap), expose via a read-heavy API with aggressive caching (CDN + Redis), and archive cold buckets to object storage. Follow-ups include handling URL normalization, deduplication of shortened URLs, and geographic segmentation. Awareness of approximate counting algorithms for streaming data, the bucketing approach for sliding windows, and the separation of the write path (stream processing) from the read path (cached ranking).

Tell me about a time you worked on a project that had significant ambiguity. How did you move it forward?

Use STAR: describe the unclear requirements or competing stakeholder interests, explain how you sought alignment (clarifying questions, written proposals, prototypes), the decision you drove, and the measurable outcome. LinkedIn's behavioral interview is described by interviewers as 'conversational and ad-hoc,' so expect follow-up probes like 'what would you do differently?' and 'how did you keep the team aligned throughout?' Align your answer to LinkedIn's transformation value — the company explicitly asks whether candidates embrace change. Comfort with ambiguity and structured thinking under uncertainty; evidence the candidate drives clarity rather than waits for it; candid reflection on tradeoffs made.

Describe the most technically impactful project you have worked on. Walk me through the key decisions and tradeoffs.

This appears in LinkedIn's hiring manager round and the project deep dive round. Choose a project where you personally drove a non-obvious architectural decision. Structure: business context, the technical options you considered, why you chose what you did, challenges you hit, and the outcome (latency, revenue, scale numbers). Interviewers at LinkedIn probe deeply — they will ask follow-on questions about any component you mention, so only describe things you can defend in detail. A typical follow-up is: 'What would you change if you were starting over?' Depth of technical ownership, ability to articulate tradeoffs clearly (not just 'we chose X'), and comfort discussing mistakes or things that did not go as planned.

Tell me about a time you disagreed with a decision made by your team or manager. What did you do?

Frame it as 'disagree and commit' executed well: you raised your concern with data or a concrete alternative, gave the decision maker a fair hearing, understood their reasoning, then either changed their mind or committed fully to the team direction. Avoid stories where you simply deferred or where you escalated unnecessarily. LinkedIn interviewers look for candidates who can influence without authority and still maintain team trust. A follow-up is often: 'How did you handle it if the outcome turned out the way you predicted?' Intellectual honesty and psychological safety; ability to advocate for a position without damaging relationships; evidence of genuine commitment once a decision is made.

LinkedIn Interview Questions

LinkedIn's SWE interview loop (IC2 through Staff) consists of a technical phone screen with 2-3 algorithm questions, a mid-process recruiter call, and an onsite of 5-6 rounds: one or two standard coding rounds, a dedicated concurrency/systems coding round (unique to LinkedIn), a system design round, a hiring manager behavioral round, and a project deep dive. As of 2025, LinkedIn has added a "coding with AI" round to the onsite loop for some roles. The process typically concludes with a team-matching call rather than a single offer; your onsite results remain valid for one year if no immediate match is found.

14 real questions across 3 rounds

Reported via LeetCode Discuss first-hand reports (2024-2025): Staff SWE June 2025 offer thread (post/6858262) explicitly names Max Stack and Minimum Window Substring (circular extension); SSE Bangalore April 2024 Offer thread (interview-experience/5039865); SSE Interview Experience thread (interview-experience/5952249) and October 2024 SSE experience mention LCA of Binary Tree; LinkedIn-tagged LeetCode discuss thread (interview-question/125144) for Find Leaves of Binary Tree; Glassdoor LinkedIn interview entry (QTN_2300794) explicitly attributes Nested List Weight Sum II to LinkedIn; jointaro.com LinkedIn-specific interview question page for Nested List Weight Sum; LeetCode discuss (post/2214902, post/5911896) confirm Delayed Scheduler as LinkedIn's concurrency-round staple asked ~6/10 times.

5 more coding questions LinkedIn 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 LinkedIn 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