June–July 2026 Amazon SDE II OA / Interviews: 10 Experiences

A practical roundup of ten firsthand Amazon SDE II and L5 reports from June and July 2026.

Coverage: Amazon and AWS · SDE II / L5 · OA, phone screen, virtual onsite, coding, LLD, HLD, Hiring Manager, Bar Raiser, and Leadership Principles · June 4–July 30, 2026

The short version

The strongest signal across these reports is that Amazon was testing much more than whether someone could finish a familiar algorithm.

1. AI-assisted debugging was a real part of the OA. Four candidates received an existing repository with failing tests. The assistant could offer hints or generate snippets, but candidates still had to read unfamiliar code, trace state, fix API behavior, and run the tests themselves.
2. A perfect OA project score was not always required. Two candidates advanced even though they did not pass every repository test. One passed four of six and still heard from a recruiter.
3. Design appeared early. One L5 candidate expected a coding problem in the phone screen and instead got a team-specific design question. Another received Leadership Principles, Validate BST, and a small system-design problem in the same screen.
4. The SDE II loop mixed implementation and design depth. Reports covered coding, Amazon Music LLD, a distributed recurring scheduler, web crawling, metrics monitoring, and a database partition split-and-merge manager.
5. Leadership Principles were not a separate final formality. Candidates saw LP questions inside coding, LLD, Bar Raiser, and Hiring Manager rounds.
6. The process after OA varied. One candidate went directly to onsite. Another moved to a phone screen. Reported follow-up ranged from a few days to roughly two weeks.
7. Ambiguity was part of the test. The hardest L5 report was not a giant algorithm. It was an open-ended partition-splitting task where the candidate spent too much time trying to discover what the interviewer wanted.

A practical preparation formula: one timed algorithm + one unfamiliar-repository debugging session + one LLD implementation + one failure-aware system design + five LP stories that can survive detailed follow-ups.

Online assessments

OA 01: Minimum circular redistribution cost + content-moderation repair

Report date: June 4, 2026 · Level: SDE II · Result: Advanced

1. Containers sit in a circle and hold different product counts. Choose one global movement direction, then redistribute products until every container has the same count while minimizing edge-crossing cost.
2. Repair an online-review service. Add and edit flows must call the existing content checker, return the correct status, persist violation counts and words, and flag repeat offenders.
3. The candidate passed four of six project tests and still advanced.

The candidate later clarified that the project used Node.js and that the built-in assistant could generate code to paste, but could not edit the repository directly.

OA 02: The same circular-cost problem + Movie DB repair

Report date: June 9, 2026 · Level: SDE II · Result: Advanced directly to onsite

1. The algorithm was the same circular redistribution problem reported on June 4.
2. The project was a Movie DB service with login, watch lists, and movie add/delete behavior. Six tests covered new and existing movies, deletion, missing watch lists, status codes, and database persistence.
3. The candidate chose Spring Java and did not pass every project test.

Recruiter follow-up came in roughly two to three days.

OA 03: Minimize the maximum grid inconvenience

Report date: July 16, 2026 · Level: SDE II · Result: Not disclosed

1. A grid already contains several service locations.
2. Add one more location anywhere in the grid.
3. Movement uses all eight neighboring directions.
4. Minimize the worst distance from any cell to its nearest location.

A commenter suggested binary-searching the answer and checking whether one new location can cover all cells that remain outside the current radius.

OA 04: House Robber II-style DSA + repository repair

Report date: July 17, 2026 · Level: SDE II · Result: Rejected

1. Solve a DSA task described as similar to House Robber II. The candidate did not publish enough detail to claim it was the exact public problem.
2. Repair a repository against failing tests with a guardrailed AI assistant.
3. Complete work-simulation choices involving LLD/HLD situations.
4. Complete Leadership Principles behavioral choices.

The candidate later confirmed that the work-simulation section was multiple-choice.

OA 05: Distinct-array rotate/flip task + six failing tests

Report date: July 21, 2026 · Level: SDE II · Result: Not disclosed

1. Given a distinct array, use only rotate and flip operations to sort it. Return the minimum operation count, or -1 if sorting is impossible.
2. Repair a repository with six failing tests and a restricted AI assistant.
3. Finish work-simulation and situational questions.

The author said the assessment took longer than expected. The exact definitions of rotate and flip were not published, so this roundup does not pretend a more specific version was reported.

Live interview question bank

Coding and data structures

1. Closest Version Date. Given a target date and dated software versions, return the nearest version date. The reported example used April 1, 2026 and selected May 3, 2026 from three candidates.
2. Postfix calculator. Evaluate an arithmetic expression written in postfix notation. No operator set or error behavior was disclosed.
3. Maximum Concurrent Processes. Given process-running intervals such as [1,3], [2,5], and [3,6], return the maximum number running at once.
4. Validate Binary Search Tree. This appeared in an L5 phone screen together with LP and design questions.
5. Range Module. The report only named the task. A comment connected it to the standard add/remove/query interval problem, but the author did not publish implementation details.
6. Koko Eating Bananas-style binary search. The candidate said the task was Koko or something similar, so treat the linked version as targeted practice rather than a word-for-word recovery.
7. Find Median from Data Stream. The report explicitly named this task.
8. Detect a Keyword Substring. Determine whether the current text contains a keyword from a supplied list.
9. Find the Root of a Directed Tree. The follow-up discussion clarified that the input was a list of parent-child edges, not an already supplied root node.

Low-level and high-level design

1. Amazon Music search, playlists, and playback. Search songs, artists, and albums; create playlists; and support play, pause, and reset.
2. Distributed recurring workflow scheduler. The candidate described it as an Airflow-like distributed CRON scheduler.
3. Web crawler. The Hiring Manager asked for an in-depth design in roughly 20 minutes.
4. Core-dump system. The report preserved only the system-design title.
5. Global metrics monitoring and alerting. This appeared in the L5 virtual onsite and was the candidate's strongest round.
6. Database partition split-and-merge manager. Split a partition above a threshold and otherwise merge suitable partitions. The original interview requirements were extremely open-ended.

Behavioral and Leadership Principles

The exact reported prompts included:

1. Tell me about a time you disagreed with your team.
2. Why are you considering changing jobs?
3. Tell me about a project that could not be completed as planned.
4. Tell me about a time AI could not generate the code you expected.
5. Tell me about a technical optimization you delivered for a customer.
6. Tell me about a technical customer problem you solved.
7. How do you use AI in your projects?
8. How do you notice and cross-check AI mistakes?

Several rounds included additional LP questions without preserving their exact wording. Those gaps stay gaps.

10 detailed candidate timelines, loops, and screens

01: OA advanced with four of six project tests

Report date: June 4 · Level: SDE II · Result: Advanced

1. Solve the circular redistribution task.
2. Repair content-moderation behavior in an existing Node.js review service.
3. Pass four of six project tests.
4. Move to a phone screen rather than directly to onsite.

This report is useful because it shows that repository understanding and partial correctness could still produce enough signal to continue.

02: OA moved directly to onsite

Report date: June 9 · Level: SDE II · Result: Advanced

1. Solve the same circular-cost problem.
2. Debug a Spring Java Movie DB project with six watch-list tests.
3. Miss some tests but receive recruiter follow-up in a few days.
4. Skip phone screen and move directly to onsite.

03: Four-round AWS virtual onsite

Report date: July 2 · Level: SDE II · Result: Rejected

Round 1

1. Solve Closest Version Date.

Round 2

1. Implement a postfix calculator.

Round 3 — Bar Raiser

1. Solve Maximum Concurrent Processes with a sweep-line approach.

Round 4 — System design

1. Design a core-dump system.

LP questions appeared in every round. The author confirmed that this was a virtual onsite and received the rejection within three days.

04: L5 phone screen mixed coding and team design

Report date: July 11 · Level: L5 · Result: Rejected

1. Leadership questions.
2. Validate Binary Search Tree.
3. A small design problem based on the hiring team's work.

The recruiter had not warned the candidate tha