Number of Islands — given a 2D grid of '1's (land) and '0's (water), count the number of islands (LeetCode 200).

Iterate every cell; on hitting an unvisited '1' increment count and DFS/BFS to mark the entire connected component visited. State the time/space complexity (O(m×n)) and discuss iterative BFS vs recursive DFS trade-offs; if the interviewer adds 'stream of coordinates' as a follow-up, pivot to Union-Find. Fluency with graph traversal on grids, clean visited-state management, and the ability to handle the follow-up (Union-Find, online stream) without prompting.

Merge Intervals — given a list of intervals, merge all overlapping ones and return the non-overlapping result (LeetCode 56). Confirmed on Glassdoor as a direct Google interview question.

Sort by start time, then greedily merge: walk the sorted list and extend the current interval's end whenever the next interval's start is less than or equal to current end. Articulate why sorting is necessary, handle edge cases (single interval, all overlapping), and expect a follow-up asking you to insert a new interval into an already-sorted, merged list. Greedy reasoning over a sorted structure, correct boundary conditions, and whether the candidate immediately identifies the simpler O(n log n) sort approach rather than an O(n^2) brute force.

Trapping Rain Water — given an elevation map as an array of non-negative integers, compute how much water it can trap after raining (LeetCode 42). Confirmed in a dedicated LeetCode Discuss thread titled 'Google | Phone Screen | Trapping Rain Water'.

Start with the O(n) space prefix/suffix max arrays approach to establish correctness, then optimize to the two-pointer O(1) space solution; at each step explain the invariant (the trapped water at index i is min(leftMax, rightMax) minus height[i]). Google interviewers specifically watch for multiple valid approaches and the ability to articulate the trade-off. Whether the candidate knows more than one approach, can reason about why the two-pointer invariant holds, and handles the edge case of empty or monotone arrays.

Design YouTube — video upload, transcoding pipeline, and global delivery to millions of concurrent viewers.

Start with scale estimates (uploads/sec, storage/day, peak concurrent streams); split into upload path (chunked upload to blob store, async transcoding workers, CDN push) and read path (CDN-first, origin fallback). Cover metadata store (video catalog DB), fan-out on upload completion, adaptive bitrate, and regional CDN failover. Google specifically expects you to explain the transcoding graph (dependency DAG of quality levels) from first principles, not just say 'use FFmpeg.' Whether the candidate can independently derive the two-path architecture, reason about codec trade-offs, handle the thundering-herd problem on viral uploads, and go deep on the transcoding pipeline rather than naming cloud services.

Design a web-scale search autocomplete system — after every keystroke, return the top-3 most frequently typed completions matching the current prefix.

Build a trie with frequency counts; for production scale, shard the trie by prefix, pre-compute top-k per node offline (nightly batch), and serve via a low-latency read tier. Discuss write path (log aggregation to MapReduce/Dataflow to trie rebuild) vs read path (prefix to shard lookup to cached top-k list). Handle prefix ambiguity, multi-language, and the cold-start problem for new queries. Trie vs inverted-index trade-off reasoning, the importance of offline pre-computation for top-k, concurrency on trie updates, and whether the candidate proactively addresses staleness and tail latency.

Design a real-time user presence system — show online/offline status and activity badges for 10M+ concurrent users, each with a name, photo, and points total. Confirmed in a LeetCode Discuss thread titled 'Google | Phone Screen | Design a webpage that can show 10M+ users status'.

Each client heartbeats a lightweight presence service every 30s; a distributed in-memory store (similar to Redis with expiry TTLs) holds the active set. Fan out status changes via a pub-sub layer to subscribers. Cover the thundering-herd problem when a popular user goes offline, the eventual-consistency window for status, and how you'd shard the presence store by user_id hash. Understanding of heartbeat/TTL semantics for ephemeral state, fan-out cost estimation, sharding strategy for the presence store, and how to handle network partition (false offline).

Tell me about a time you faced a conflict with a teammate or stakeholder. How did you handle it, and what was the outcome?

Use STAR (Situation, Task, Action, Result); focus the Action on specific steps you took to understand the other person's perspective before advocating your own position. Google is looking for 'emergent leadership' not positional authority — show you resolved it through data and persuasion, not hierarchy. End with the concrete outcome and what you'd do differently. Interpersonal maturity, ability to de-escalate without losing conviction, and whether the candidate demonstrates intellectual humility (updating their view when shown data) vs defensiveness.

Describe a time you had to make a significant decision under significant ambiguity, with incomplete data. What did you do?

Establish why waiting for perfect information was not an option, identify the minimum viable signal you gathered, explain your decision framework, and quantify the impact. Google tests 'comfort with ambiguity' as a core Googleyness trait — explicitly name the uncertainty, show how you bounded the risk, and reflect on whether the decision held up. Structured thinking under uncertainty, a bias toward action over analysis paralysis, and honest post-mortem self-reflection (including if the outcome was mixed).

Give me an example of when you went above and beyond what was expected of you on a project or for a teammate.

Pick a story where the extra effort was driven by intrinsic motivation or team impact, not recognition-seeking. Quantify the impact ('reduced deploy time by 40%', 'unblocked 3 engineers for a week'). Google's Googleyness rubric penalizes self-promotion; frame it as 'I saw the team needed X so I did Y' rather than 'I wanted to stand out.' Genuine collaboration instinct, concrete impact metric, and absence of credit-hoarding — interviewers watch for 'we' vs 'I' framing.

Design a YouTube Shorts recommendation system — given a user's watch history and real-time session signals, rank a candidate set of short-form videos to maximize watch time and satisfaction.

Two-stage architecture: candidate generation (embedding-based ANN retrieval from a user tower plus item tower trained with a contrastive or two-tower model) followed by ranking (a deep neural ranker with dense features: watch duration, skip rate, freshness, user context). Discuss training data (implicit feedback, position bias correction), serving latency constraints (under 100ms), and how you'd detect and mitigate filter-bubble effects. Fluency with two-tower retrieval vs ranking separation, awareness of implicit feedback bias, real-time feature serving architecture, and the ability to reason about offline metrics (AUC, NDCG) vs online metrics (watch time, satisfaction).

Google Interview Questions

Google's SWE loop (L3–L5) runs 4–5 onsite rounds (typically 2–3 coding, 1 system design at L5+, 1 Googleyness/Leadership behavioral), preceded by an OA and 1–2 phone screens; as of early 2025 Google re-introduced at least one in-person round and added the Google Hiring Assessment (GHA) personality/situational-judgment screen for all levels to combat AI-assisted cheating.

22 real questions across 4 rounds

Reported via LeetCode Discuss 2024 Google compilation thread (6185127, 6469509); individual L4 Oct 2024 Bangalore experience post (5958167) and matching 'Count Good Subarrays' thread (5955526); New Grad 2024 post (4603668); dedicated LeetCode Discuss threads 'Google | Phone Screen | Trapping Rain Water' and 'Google | Phone | Find Median from Data Stream' and 'Google onsite: Serialize & Deserialize Binary Tree with String field'; Glassdoor Google SWE tag confirms Merge Intervals and Number of Islands.

7 more coding questions Google 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 Google 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