Amazon Interview Questions — Page 105

Amazon·Software Engineer·Onsite - Coding / Algorithms

Jun 2026

Amazon SWE coding round, two questions back to back. The second was a verbal-only follow-up on prefix sums, no actual coding needed.

Given an array of integers, find a subarray whose elements sum to zero. Explain your approach (no coding required).

“They just wanted the explanation, which felt like a relief at first.”

View Post

Amazon·Software Engineer·Technical Phone Screen

Jun 2026

Amazon coding round, one algorithmic problem the whole session. The problem looked like a string manipulation warmup but turned into something way more involved once you actually think about the constraints.

Given a string, for each prefix of that string compute the maximum number of equal-length contiguous blocks you can split it into such that every block has the same character frequency multiset.

“I stared at the example with 'ABBA' for a solid minute before I even understood what 'same frequency multiset' meant in context.”

View Post

Amazon·Software Engineer·Technical Phone Screen

Jun 2026

Amazon SWE coding round with a greedy string-building problem that looked approachable until the state-flip mechanic made it way more annoying than expected. The heap-with-lazy-admission pattern was the key insight I didn't fully have going in.

Given an array of positive integers and a binary state string of the same length, build a result sequence of length m by repeatedly picking the largest currently available element (available means state[i] == '1'). After each pick, any '0' in the state that sits directly to the right of a '1' flips to '1', unlocking more elements. Return the lexicographically largest result sequence.

“I spent the first few minutes thinking this was just a straightforward greedy max problem and almost missed the flip mechanic entirely.”

View Post

Amazon·Software Engineer·Technical Phone Screen

Jun 2026

Amazon SWE coding round, one algorithmic question that's basically a generalized version of longest consecutive sequence with a gap parameter k. Not the hardest thing I've ever seen but the edge cases and the duplicate ambiguity caught me a little flat-footed.

Given an integer array and a positive integer k, find the length of the longest chain of distinct values from the array where every adjacent pair in the chain differs by at most k.

“I went straight to the hashmap trick from the classic consecutive sequence problem and then realized mid-explanation it doesn't cleanly port over.”

View Post

Amazon·Software Engineer·Technical Phone Screen

Jun 2026

Amazon SWE interview with a palindrome coding problem. Not much to say, it was a single algorithmic question and I got through it.

Find all palindromes in a given input.

“Pretty classic problem but I fumbled the edge cases a bit.”

View Post

Amazon·Software Engineer·Onsite - System Design / Architecture

Jun 2026

Amazon system design round focused entirely on building a GitHub-scale developer platform. Dense question, lots of ground to cover, and I definitely felt the time pressure trying to hit every subsystem they wanted.

Design a GitHub-style developer community platform covering user profiles, repositories with code, issues, and pull requests, follow/star relationships, an activity feed, and scaling for high throughput. Walk through sharding, read/write paths, caching, feed fan-out, search, code blob storage, key API endpoints, and the data model.

“This is basically five questions rolled into one.”

View Post

Amazon·Software Engineer·Onsite - Behavioral / Leadership

Jun 2026

Amazon SWE interview with a scenario-based question about handling a teammate who wants to skip integration testing to hit a PM deadline. Pretty situational, felt more like a judgment test than a technical one.

Your teammate wants to skip integration testing on a purchase feature to meet a PM deadline. Rank the following responses from most to least effective: reminding him that automated testing beats manual, looping in your manager or Amber, pushing back and offering to help with testing yourself, suggesting a compromise meeting, or just agreeing to skip the tests.

“I underweighted the 'loop in manager' option because it felt like tattling.”

View Post

Amazon·Software Engineer·Onsite - Behavioral / Leadership

Jun 2026

Amazon behavioral round for a software engineer role, two questions back to back that were clearly related. Both felt like they were probing the same underlying thing: how you handle not knowing stuff and whether you're thoughtful about your own blind spots.

Tell me about a time you had to get up to speed on an unfamiliar technology or domain quickly to ship something. How did you approach the learning, what did you use to ramp up, how did you know you knew enough, and what happened?
How do you actually use AI coding tools in your day-to-day work? Give real examples of where they help you and where you don't trust them. What do you do to review or sanity-check what they produce?

“The 'how did you know you knew enough' part is what trips people up.”

View Post

Amazon·Software Engineer·Online Assessment (OA)

Jun 2026

Amazon SWE coding round where they handed me a broken repo and told me to fix it. No algorithm puzzle, just real debugging work on a password reset flow with multiple layers of issues stacked on top of each other. More stressful than a LeetCode problem honestly because the bugs weren't obvious at first glance.

You are given a partially broken codebase for a password reset flow. The flow has three steps: generate a verification code, verify it, then reset the password. Multiple bugs are present. Find and fix them.

“The first bug was almost embarrassing once I found it.”

View Post

Amazon·Machine Learning Engineer·Technical Phone Screen

Jun 2026

Deep technical interview for an ML engineer role on a reasoning model team at Amazon, focused almost entirely on RL post-training with GRPO. The interviewer clearly wanted hands-on familiarity, not textbook definitions. Five heavy topics back to back with follow-ups that assumed you'd actually run these training jobs before.

Explain GRPO: what problem does it solve compared to PPO, why does it drop the critic network, and what does the advantage formulation look like?
Walk through the parallelism strategies for training a large model at scale and explain how DualPipe works and how nodes communicate with each other.
Describe Multi-head Latent Attention. What problem does it solve, how does the KV compression work, and why is the RoPE component handled separately?
How would you design the reward function for RL training of a reasoning model? Make the case for rule-based vs. learned reward models and for process vs. outcome rewards.
What failure modes show up in practice when training with GRPO, and how do you diagnose and fix them?
GRPO normalizes advantages by the group standard deviation and divides the per-sample loss by response length. Where does each introduce bias, and how would you remove it while keeping the critic-free structure?
You see reward going up steadily during training but held-out accuracy is flat or declining. How do you diagnose whether it's reward hacking or a real train-eval gap, and what do you change first?

“This is where I spent the most time and I think I did okay but fumbled the math a bit under pressure.”

View Post