Amazon SDE Intern Interview Experience (2026) - Concatenated Words & Resume Deep Dive VO, Offer

• "What's the most complex project you've worked on? How did you weigh options?"
• "Tell me about a project that went really wrong. What did you take away?"
• "What would you do if you realized partway through that your approach wasn't the best path forward?"

The second question is the Failure question in disguise, and the third is the Bias for Action self-correction probe. The worst thing you can do on the second is pick a project that was externally killed. They want a technical decision you personally made that went wrong. The cleanest version is "I picked sync over async for this pipeline because I underestimated traffic burstiness. Two weeks later we had to migrate. Here is what I changed in how I estimate load now."

This round was purely behavioral, no code. I kept each answer around 3 minutes. The interviewer visibly relaxed when I hit the lessons-learned beat inside 2 minutes.

Round 2: BQ plus Coding, Concatenated Words

The second round shifted to a more LP-heavy BQ block followed by a single coding problem. The BQ portion was roughly 20 minutes, the coding portion about 35.

BQ probes this round:

• Failure and what you did about it (different angle from round 1, they want a second data point)
• A disagreement with a manager or lead, how you resolved it, whether you won or conceded
• A time you took ownership of something outside your direct responsibility

The disagreement question is the one where most interns over-index on "I convinced them." Amazon actually wants to see that you can dissent, argue well, and then either commit when overruled or escalate cleanly. Saying "I realized they were right and changed my mind" is a valid answer if the story supports it.

Problem: Given a list of words, return all words that can be formed by concatenating at least two other words from the same list.

This is Word Break extended to a filtering problem. My approach:

1. Put every word in a hash set for O(1) membership.
2. For each candidate word `w`, remove `w` from the set temporarily so it cannot use itself as a component.
3. Run Word Break on `w` using the remaining set: `dp[i]` is true if `w[:i]` can be built from set words, with `dp[0] = true`. Transition: for each split `j < i`, `dp[i] = dp[j] and w[j:i] in set`.
4. Put `w` back into the set and move to the next candidate.

Follow-ups the interviewer pushed on:

• Complexity analysis. I said O(N * L^2) where N is total words and L is max word length, since Word Break per word is O(L^2) with set lookup.
• Pruning: if any word cannot be broken into at least two components, we need to reject even if `dp[n]` is true. The protection above comes from removing `w` from the set first, which forces any valid decomposition to use other words.
• What if words share long prefixes? I mentioned a trie would be a natural swap for the inner loop, reducing substring lookup overhead. I did not implement it, just described.

I also verified one edge case out loud: empty string. If the list contains an empty word, it should be skipped, since an empty word being "concatenated" is a degenerate case.

Practice it: [[problem/171?company=3|Decode Ways]]

Practice it: [[problem/148?company=3|Longest Substring Without Repeating Characters]]

Result

Recruiter reached out 4 business days later with a verbal offer. Standard Amazon intern comp for a US tier-1 city. Asked for 10 days to decide, they granted 14.

Tips

1. Prep two distinct Failure stories. Amazon asks the failure question in both rounds, phrased differently. If you only have one story, the second time you recycle it, the interviewer writes you down as having limited self-awareness. Mine were one technical decision I regretted and one cross-team communication I botched.
2. For disagreement questions, do not always "win." A story where you respectfully lost the argument, committed to the decision, and the outcome validated your manager is stronger than "I convinced them." Amazon specifically tests whether you can disagree and commit.
3. Concatenated Words is hot this cycle. Cover Word Break first, then the concat variant. The trap on concat is forgetting to exclude the word itself from the lookup set, which causes single-word "decompositions" to falsely pass.
4. Narrate while you code. The interviewers explicitly noted they wanted to "hear my thinking." On Concatenated Words, I called out the self-exclusion trick before writing the `remove` line, and that earned a nod. Silent coding gets you penalized even if the code is correct.
5. Aim for 3-minute STAR, not 5. Interns often pad stories. A crisp Situation-Task-Action-Result in 3 minutes leaves room for 2 or 3 follow-up questions, which is where the interviewer actually evaluates you. Pad and you will only get one follow-up, which is worse signal.
6. Round 2 BQs dig harder than round 1. The first round is warm-up. The second round asks the same LPs at a deeper level. Have one shallow and one deep version of each of your 6 stories ready. The interviewer chooses which level to probe.