Amazon Intern 2-Round VO Interview Experience (2026) - Word Ladder BFS & Concatenated Words DP, Offer

The second one maps cleanly to Deliver Results. What the interviewer actually cared about was not the story itself but whether I could explain the decision I made, why I made it, and what the measurable outcome was. I opened my answer with the deadline: "We had 3 weeks between the feature spec landing and the launch date, and half the team was on paternity leave." That context-first framing earned an immediate "okay, go on."

Problem: Given beginWord , endWord , and a word list, find the length of the shortest transformation sequence where each step changes exactly one letter and every intermediate word is in the list.

Classic Word Ladder. My approach:

1. Put the word list in a hash set.
2. BFS starting from `beginWord` with distance 1.
3. At each dequeue, for every position i in the word and every letter `a..z`, construct the candidate. If the candidate is `endWord`, return distance + 1. If it is in the set and unvisited, enqueue with distance + 1 and mark visited.
4. Return 0 if BFS drains without reaching `endWord`.

Time complexity: O(N L 26) where N is word list size and L is word length. The interviewer asked about bidirectional BFS. I described it verbally, search from both ends until the frontiers meet, roughly halving the exponent, but didn't rewrite the code. They accepted that.

The detail that tripped the candidate sitting next to me in the prep cohort: skipping the visited check, which causes re-enqueueing and the grader times out on the 5000-word test set.

Round 2: BQ plus Concatenated Words DP

Second interviewer had a different tone, more pointed, wanted specific failure stories and specific pushback stories.

BQ probes:

• Tell me about a time you failed and what you learned
• Describe a situation where you disagreed with a leader and how you handled it

The disagreement question is where I had over-rehearsed. The interviewer interrupted my first pass and asked "what did you actually say to your manager in that meeting?" I had to recover and give the literal sentence I used. The lesson: prep your BQs down to the dialogue level, not the narration level.

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

This is Word Break with a self-exclusion wrinkle. My solution:

1. Insert all words into a hash set.
2. For each word `w`, temporarily remove it so the DP cannot use `w` itself.
3. Compute `dp[i]` = can `w[:i]` be built from remaining set words. Base `dp[0] = true`. Transition: `dp[i] = any(dp[j] and w[j:i] in set for j in 0..i-1)`.
4. If `dp[n]` is true, append `w` to the result. Put `w` back.

Two correctness traps I called out explicitly before coding:

• Not removing `w` from the set means a single-word "decomposition" passes, violating the "at least two words" requirement.
• Empty-string edge case should be skipped rather than counted.

Complexity: O(N L^2) per word, with set lookup O(L) for the substring hash. The interviewer asked about a trie variant for the inner substring check. I talked through how a trie rooted at position j would let you walk characters and check valid break points inline, reducing substring construction overhead, but did not implement it.

Result

Verbal offer came 6 business days after VO. Standard intern comp, Seattle office. I accepted within the response window since my alternatives were in the same bucket and Amazon gave me the quickest turnaround.

Tips

1. BFS template at muscle-memory speed. Word Ladder is specifically the test of whether you can write `deque + visited + distance` without thinking. Drill it until you can type it in under 6 minutes, then spend the rest of the round on edge cases and the bidirectional follow-up.
2. Concatenated Words: remove-then-restore is the crux. Every failed attempt on this problem in my prep cohort came down to forgetting the self-exclusion step. Frame it as "the word itself can never be a component," then coding becomes obvious.
3. Prep BQ dialogue, not narration. Amazon interviewers will interrupt and ask "what did you actually say?" or "what was the exact tradeoff you wrote in the doc?" Have one literal sentence ready per story.
4. For deadline stories, lead with the constraint. Opening line should be the number of weeks, number of engineers, or scope of the deliverable. Amazon behavioral evaluators are specifically trained to look for "quantified pressure" in Deliver Results answers.
5. Two rounds back-to-back is a stamina problem. Don't skip lunch. Don't schedule the VO at 5pm if you can avoid it. Mental fatigue in round 2 is when your Failure story drifts into excuse-making territory, and that is a direct "no hire" signal.
6. Reuse LP stories across rounds, but rotate angles. Using the same project across two rounds is fine, but the emphasis must be different. Round 1 framed mine around "Deliver Results under deadline." Round 2 framed the same project around "what I would do differently" for the Failure question.

The intern loop is beatable. The pool is large and the bar for interns is about showing you can learn fast, communicate well, and write standard algorithms cleanly. Put a week of focused BFS, DP, and BQ prep into it and you will clear the loop.