Amazon Interview Experiences

1. 📝

   Online Assessment (90 min)

   I'd expected the OA to be the throwaway filter round, the part everyone treats as a formality. It wasn't. Ninety minutes for two problems sounds generous until the second one turns out to need a data structure I hadn't touched in months — I ended up sketching the state transitions on a scrap of paper beside my laptop, the old-fashioned way, before trusting myself to type it. The behavioral MCQs at the end felt almost jarring after that — dozens of forced-choice scenario questions, no time pressure, but you can tell Amazon is quietly screening for something with those too.

2. 📞

   Phone Screen (elimination round)

   My interviewer for this one — I'll call her Priya, she introduced herself as a senior SDE-2 — started with almost no small talk, which threw me a little. The first problem, sliding-window anagrams, I'd genuinely drilled before, so that part went smoothly. The second was the diagonal-connectivity island question, and I made the classic mistake out loud: I described my flood-fill using only the four cardinal directions. She didn't correct me immediately. She just asked, 'are you sure that grid in your second example only has one island?' — and let me sit with it for a few seconds before I caught my own bug. That's the moment I actually remember from the whole round, more than the coding itself.

3. 💻

   Loop Round 1 — DSA

   This interviewer cared less about whether I reached the right answer and more about how I got there. I opened with the brute-force O(n·maxHeight) scan, said out loud that it would time out on large inputs, and he immediately leaned forward — 'good, now what's monotonic about this problem?' That single question was the whole hint I needed to see it as binary-search-on-the-answer instead of a linear scan. He spent the last ten minutes purely on complexity trade-offs, barely glancing at the code itself.

4. 🧩

   Loop Round 2 — LLD

   I'd rehearsed LLD prompts enough to know the shape of a good answer, but this interviewer — Priya again, she sat in on two of my rounds — refused to let me skip requirements gathering. Every time I moved toward drawing a class, she'd ask another clarifying question: 'what happens with recurring bookings?', 'do we need a buffer between meetings?' By the time I actually started naming classes, I'd spent almost fifteen minutes just on requirements, and the design itself came out cleaner for it. The follow-up that stuck with me: 'why Strategy for the conflict check and not just an if-else?' — and I had to actually defend the choice, not just name it.

5. 🎯

   Loop Round 3 — Bar Raiser

   The Bar Raiser's whole opening move was silence. He gave me one sentence — 'design a package locker system' — and then just waited, arms folded, watching whether I'd start designing immediately or start asking questions. I'd read enough about Bar Raisers to know that trap, so I spent the first five minutes purely on clarification: single building or a network of lockers, one carrier or many, what happens when a package sits unclaimed for a week. He didn't say much, but I noticed him nodding at the unclaimed-package question specifically — that turned out to be the crux of the whole design.

6. 🏗️

   Loop Round 4 — HLD

   By this point in the day I was running on adrenaline more than energy, four rounds deep. The HLD prompt — keep a user's settings in sync across devices — sounded almost relaxing after the locker system, until the interviewer asked the question that actually mattered: 'what if a device comes back online after being offline for a week?' I hadn't planned for that case in my initial design, and I said so honestly instead of bluffing an answer — then thought out loud through rebasing local changes on top of the latest server state before pushing anything back. He seemed more interested in watching me reason through the gap live than in whether my first-pass design had already covered it.

Key takeaway: The full loop spanned about four months; prepare DSA, LLD, HLD, and Leadership Principles as four separate tracks rather than blending prep, and lean on real product experience when designing LLD systems.

1. 📝

   Online Assessment

   I'd cleared plenty of timed contests before, so the 90-minute window didn't scare me — until the second problem turned out to be a matrix question I'd only half-practiced. I burned almost 20 minutes just re-deriving the recurrence on paper before typing a single line. What actually saved the round wasn't cleverness; it was forcing myself to write the brute-force version first, get it passing on the sample, and only then optimize. The Work Style Assessment afterward felt almost like a breather in comparison — a long questionnaire, no coding, just picking the response closest to how I actually behave under pressure.

2. 🔒 3 more rounds — continue this journey in the full kit

1. 💻

   Round 1 — Coding

   The interviewer, I'll call him Kabir, opened with an easy warm-up and let me talk through my approach out loud before I typed anything — I later realized that was deliberate, he was watching how I think, not just whether I got the syntax right. The second problem looked like a plain BFS at first glance, but he'd added a combination-lock-style twist to the state space. I almost modeled it as a simple graph and would have missed an entire class of valid moves; catching that myself, out loud, mid-explanation, seemed to matter more to him than getting to the answer fast.

2. 🔒 3 more rounds — continue this journey in the full kit

1. 📞

   Phone Screen

   This was over video, no whiteboard, just a shared editor — which somehow made it feel more exposed, not less. I recognized the prefix-sum pattern almost immediately, which in hindsight made me a little overconfident; I started typing before I'd said out loud what seen[running - k] was actually tracking. The interviewer stopped me gently — 'walk me through what that hashmap represents before you write the loop' — and that one course-correction is probably what got me through the follow-ups that came after.

2. 🔒 4 more rounds — continue this journey in the full kit

1. 📝

   Round 0 — Online Assessment

   This OA had a section I hadn't seen in older prep material — an AI/tooling-fluency component alongside the usual DSA question. It wasn't hard exactly, just unfamiliar, and unfamiliar under a timer is its own kind of hard. I remember consciously slowing down on the instructions instead of skimming them, because the format itself was new and I didn't want to lose points to a misread requirement rather than a wrong algorithm.

2. 🔒 4 more rounds — continue this journey in the full kit

1. 💻

   Round 1 — Technical (DSA)

   I got the core algorithm out fast — almost too fast — and the interviewer noticed. Instead of moving on, she spent the remaining twenty minutes throwing edge cases at my own solution: an empty array, all-negative numbers, a target of zero. I hadn't actually dry-run those myself before declaring I was 'done', and catching two small bugs live, in front of her, felt worse than not knowing the algorithm at all. In hindsight that was probably the actual test — not whether I could solve it, but whether I'd trust my first answer or actually stress-test it.

2. 🔒 2 more rounds — continue this journey in the full kit