Number of Islands - given a 2D grid of '1' (land) and '0' (water), count the number of distinct islands (connected components).

Recognize it as connected-components on an implicit graph; flood-fill each unvisited land cell with BFS or DFS, marking visited in place to avoid revisits. State O(rows*cols) time/space, then field follow-ups like counting island area, diagonal adjacency, or the streaming 'Number of Islands II' union-find variant. Grid-as-graph modeling, clean BFS/DFS with visited tracking, correct boundary handling, and complexity reasoning. This is one of Amazon's single most-asked problems (freq ~3.78), so fluency under follow-ups matters.

Design and implement an LRU Cache supporting get(key) and put(key, value) in O(1).

Combine a hash map (key -> node) with a doubly linked list ordered by recency; move-to-front on access, evict from the tail at capacity. Walk through node-splice helpers and edge cases (update existing key, capacity 1). Be ready to extend to LFU or to make it thread-safe/TTL-based for senior loops. Picking the right two data structures together, getting the O(1) invariant right, and clean pointer manipulation. Amazon uses this as a calibration/design-lite problem (freq ~3.11) and it shows up in Glassdoor SDE-II reports.

Rotting Oranges - in an m x n grid, fresh oranges (2) rot adjacent fresh ones each minute; return minutes until none are fresh, or -1 if impossible.

Multi-source BFS: seed the queue with all initially rotten cells, BFS level-by-level counting minutes, decrement a fresh counter. Return -1 if fresh remain. Contrast with DFS to show why BFS gives the minimum-time/level answer. Identifying multi-source BFS, level-by-level time tracking, and the unreachable-fresh edge case. A top-5 Amazon-tagged problem (freq ~3.63).

Design a URL shortener (TinyURL): given long URLs, generate short codes and redirect, at large scale.

Clarify scale, read/write ratio, and code length; pick an ID strategy (counter+base62, or hashing with collision handling), design the KV store and redirect path, then layer caching, custom aliases, expiry, and analytics. Discuss read-heavy caching and DB partitioning. One of the most commonly reported SD prompts across FAANG including Amazon. Evaluates requirements gathering, the encoding/ID-generation tradeoff, data modeling, and cache/scaling reasoning end-to-end.

Design a rate limiter (e.g., limit a client to N API calls/sec across a fleet).

Pick an algorithm (token bucket / sliding window) and justify it; decide local vs. distributed counters with a shared store like Redis, and handle clock/atomicity (Lua/atomic ops). Be ready for the standard concurrency follow-up - many threads hitting the counter at once - and reason about race conditions and per-key (not whole-map) locking. Algorithm choice with justification, distributed-state and concurrency correctness, and graceful behavior at the limit. A reported Amazon LLD/SD prompt; interviewers typically push on the multithreaded thread-safety follow-up.

Manage capacity for a network of Amazon delivery lockers: allocate mixed-size locker slots to incoming packages to minimize cost per package.

Model slot sizes and package sizes, the assignment/bin-packing policy (smallest-fitting slot, fallbacks when full), reservation/expiry of slots, and the pickup-code flow. Discuss data model, concurrency on slot allocation, and what to do when a location is full. A real last-mile Amazon problem confirmed near-verbatim by the Exponent guide. Tests allocation-policy design, handling constraints (big-in-small not allowed), failure/overflow handling, and pragmatic data modeling over buzzword scaling.

Tell me about a time you took on something significant outside your defined responsibilities. (Ownership)

Use STAR: pick a story where you saw a gap nobody owned, took it on without being asked, and drove it to a measurable outcome you stayed accountable for. Emphasize long-term thinking ('never say that's not my job') and what you'd do again. Ownership is among the LPs Amazon weights most for SDEs of every level: acting autonomously, owning outcomes end-to-end, and not waiting to be told. Vague 'we' answers without personal accountability fail.

Tell me about a time you had to make a decision with incomplete data or information. (Bias for Action / Are Right, A Lot)

Show a reversible, time-pressured call: state what data you had, the calculated risk, how you moved fast, and how you de-risked (small bet, fast feedback). Close with the result and how you corrected if wrong. Bias for Action - comfort with speed under uncertainty and calculated risk-taking, paired with judgment. Interviewers probe whether you distinguished one-way from two-way doors.

Tell me about a time you disagreed with your manager or a senior leader. What did you do? (Have Backbone; Disagree and Commit)

Pick a real, respectful disagreement backed by data; explain how you escalated/advocated, then - crucially - committed fully once the decision was made, even if it went against you. End with the outcome and the working relationship preserved. Backbone plus the 'commit' half: principled pushback without being political, then wholehearted execution. Stories with no real conflict, or where you stayed bitter, score poorly.

Amazon Interview Questions

Amazon's SDE loop is an online assessment (2 coding problems + a Work Simulation + a Work Style survey) followed by a 4-5 interview onsite. Every interview is roughly half coding/design and half Leadership Principles behavioral, each interviewer owns specific LPs, and a Bar Raiser from another team has veto power; SDE2+ adds a dedicated system design round.

23 real questions across 3 rounds

Reported via LeetCode Amazon company tag + krishnadey30/LeetCode-Questions-CompanyWise amazon_1year.csv (frequencies verified to match), DesignGurus Amazon list, and Glassdoor Amazon SDE/SDE-II reports (2025-2026). All problems below sit at the top of the Amazon-specific frequency lists; ~75-80% of asked problems are Medium.

7 more coding questions Amazon 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 Amazon 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