Hi! This is the begnning of another chapter 🚀

A roundup of 9 Amazon engineering interview experiences from September 2026. I marked the ones that were confirmed to be SDE II and left the level blank when it wasn’t mentioned. But they are belong to full-time with 3 - 5 yoe

Coverage: Five reports explicitly identified as SDE II, Software Engineer II, or Amazon II, plus four full-time reports whose exact level was not explicitly mentioned.

The reports covers OA, phone screens, coding, LLD, HLD, Hiring Manager, Bar Raiser, and Leadership Principles. Locations include the United States, Bangalore, and undisclosed locations.

The short version

The clearest September signal is all the probing and follow-ups after the candidates' first BQ answer.

A practical preparation formula: unfamiliar-repository debugging + clean medium-level coding without an IDE + one complete LLD implementation under time pressure + failure-aware system design + LP stories that survive three layers of evidence checks.

Online assessments and early screens

OA 01: Drone delivery around a loop + AI repository debugging

Report date: September 2026 · Level: Explicit SDE II · Outcome: Advanced

1. A drone delivered packages clockwise around a loop. Travel time was given between each neighboring pair of points. The task was to compute the time needed to complete all deliveries along the route. The candidate recognized a prefix-sum approach.
2. An AI-coding task supplied a complete project with bugs. The candidate reported that the AI was not very useful and that reading the function definitions in the README was enough to identify the required fixes.
3. The candidate received a phone-screen invitation the next day, but that position was filled before the screen took place. Amazon then offered three other openings to choose from.

OA 02: AI-assisted repository debugging, prompt withheld

Report date: September 13, 2026 · Level: Explicit SDE II · Result: Offer

1. Complete an AI-assisted repository-debugging task.
2. Solve one DSA problem.

The candidate did not remember the DSA prompt and did not disclose the repository, language, allowed AI tools, bug count, scoring, or hidden tests.

OA 03: Binary search on answer + Python/Django debugging

Report date: September 9, 2026 · Level: Explicit SDE II · Location: Bangalore · Experience: 3.3 years

1. Solve an undisclosed problem whose key was recognizing binary search on the answer.
2. Debug an existing Python/Django project.
3. An AI assistant was available but deliberately did not reveal the bug or exact fix, so most debugging remained manual.

Live interview question bank

Arrays, graphs, and stateful objects

1. Custom BFS that grows into multi-source BFS. The interviewer progressively added constraints. The candidate completed the implementation but silently left one known edge case uncovered.
2. Find the safest path in a grid. The candidate used multi-source BFS to compute distance from danger, binary-searched the safeness threshold, and used DFS for feasibility. The interviewer then asked whether it could be optimized further.
3. Count stepping numbers in a range. The candidate solved it with BFS. One follow-up was asked but not disclosed.
4. Capacity to ship packages within D days. Reported as a standard binary-search-on-answer question.
5. Container With Most Water → Trapping Rain Water. The second problem was an explicit follow-up to the first, with a deeper discussion of the intuition.
6. Cheapest Flights Within K Stops. Treated as a graph/BFS minimum-weight problem.
7. Valid Palindrome-style screen. The discussion focused on the approach, edge cases, and time/space complexity. The exact variant was not preserved.
8. LeetCode 146, LRU Cache. One candidate solved the opening problem quickly and then received two undisclosed follow-ups.
9. Latest-timestamp key-value object. Support set(timestamp, key, value) and get(key), where get returns the value associated with the greatest timestamp written for that key. The report compared the object to LRU but did not mention capacity or eviction.
10. LeetCode 895, Maximum Frequency Stack. The candidate had about 15 minutes left, recognized a stack-of-stacks direction, but supplied an O(log n) approach and did not complete the code.
11. Knapsack-style Bar Raiser problem. A greedy approach was challenged. After hints and discussion, the candidate moved toward DP but ran out of implementation time.
12. Vehicle-inspection registration scheduler. Vehicles require registration every six months. Registration is closed on holidays, initially runs only from 10 a.m. to 4 p.m., and uses one-hour slots. Given n waiting vehicles, distribute registrations as evenly as possible.

• The candidate sketched separate slot-generation and scheduling functions.
• Follow-up: change the daily registration window to 11 a.m.–3 p.m.

Low-level design

1. Return drop-off reservation system. A customer reserves a slot at a nearby store and physically returns a package there. The candidate modeled entities, interfaces, booking flow, and store availability.
   • Follow-up: find nearby eligible stores efficiently.
   • The candidate mentioned geosharding; the interviewer hinted at the way ride-sharing systems locate nearby drivers.

High-level and distributed-system design

1. Distributed rate limiter. Compare fixed window, sliding window, and token bucket; then discuss Redis, locking, concurrency, multiple application instances, consistency, and trade-offs.
2. Robot restaurant. Design how many robots in a restaurant should coordinate and work. The original prompt was deliberately broad, and the candidate reported many follow-up questions without preserving each one.
3. Multi-device configuration synchronization. Discuss APIs, data model, versioning, propagation, retries, idempotency, offline devices, conflict resolution, and consistency. The interviewer kept adding failure scenarios.
4. Amazon inventory management. Key follow-up: only one unit remains and two users order concurrently. Prevent both orders from succeeding. The discussion covered race conditions, concurrent updates, inventory consistency, overselling, locks, and transactions.
5. Kafka-based file watcher for millions of files. Design a file-watcher system that detects changes across millions of files and uses Kafka to distribute events. The source gives the system shape but not the full requirements or follow-up sequence.
6. Burst-tolerant notification ingestion. Design a microservice that receives notifications with low latency during sudden bursts. The candidate reported that clarification questions were answered with “it is up to you,” so the exact target contract remained broad.
7. Four-round onsite pattern. One candidate reported that coding stayed near LeetCode medium but was followed by extensions, while system design repeatedly drilled into “why,” scale, consistency, and failure handling.

Behavioral questions and follow-up audit

The list below separates exact candidate-reported prompts from LP labels and generic advice.

Exact or closely preserved main questions

1. Tell me about a time you missed a deadline.
2. Tell me about a project you started on your own initiative.
3. Tell me about a project with large impact.
4. How did you build the AI-agent project on your resume?
5. How many people used it?
6. Which departments did you work with?
7. What was your biggest mistake?
8. Did you proactively improve a company process?
9. Tell me about conflict with others and how you handled criticism.
10. Tell me about something new you learned and applied at work.
11. What did you personally do on the project that created the greatest benefit for the company?
12. What did you do for a customer that made them especially satisfied?
13. Tell me about a time you needed to take responsibility or ownership.
14. Tell me about a time you faced a challenge.
15. Under a tight deadline, was there a project you could not complete because of time?
16. Tell me about a problem that required you to dive deep until you found the issue. The candidate described this as essentially a debugging story.
17. Tell me about a conflict.
18. Tell me about a tight timeline.

Follow-ups actually reported

1. The missed-deadline, self-initiated-project, and high-impact-project questions all had follow-ups, but their exact wording was not preserved.
2. The ownership/challenge interview followed up on AI usage, impact, number of users, the manager’s reaction, and related measurable outcomes.
3. The AI-agent resume discussion explicitly drilled into implementation, adoption, and cross-department use.
4. One onsite candidate said the three questions about new learning, company benefit, and customer satisfaction had no follow-ups.

LP labels without the question wording

• Learn and Be Curious
• Ownership
• Dive Deep
• Invent and Simplify
• Customer Obsession
• Deliver Results
• Earn Trust
• Have Backbone; Disagree and Commit
• Bias for Action

Several reports only named the principle or said “standard Amazon LPs.” Those sources do not support inventing a main question or follow-up.

Continue to Part 2

Part 2 contains all 9 detailed candidate timelines, with FastPrep practice links repeated where useful, followed by the evidence-weighted priorities and final checklist.