很多人一提到 Amazon OA，第一反应就是：
“不难吧？刷点 LeetCode + 看下 LP 就行？”但这几年我们在 Programhelp 实际陪跑了大量 Amazon OA 之后，发现一个很明显的现象：真正被刷掉的人，大多不是“不会写题”，而是没理解 Amazon OA 的真实筛选逻辑。
这篇文章，我会从真实候选人的 OA 反馈 + 我们长期总结的数据出发，完整拆解 Amazon OA 的流程、题型分布、高频考点，以及最容易被忽略、但杀伤力极大的细节。
Amazon OA 整体流程概览（SDE / New Grad 通用）
以近一年 SDE / NG 岗位为例，大部分 Amazon OA 由以下几部分组成（顺序可能略有调整）：
- Online Coding Assessment
- 通常 2 道算法题
- 平台多为 HackerRank
- 时间 70～90 分钟不等
- Work Style Assessment（性格 / 行为测评）
- 强相关 Amazon Leadership Principles
- 没有“标准答案”，但有明确的“雷区”
- 部分岗位会加：Work Simulation / Debugging / SQL
- 尤其是 SDE Intern、部分 Team 特定岗位
Notice:
OA 本身就是第一轮筛选，不是“走流程”。
很多简历不错的候选人，直接卡在 OA，没有 VO。
Amazon OA 真题分享
MySQL 挑战
Problem-solving ideas
- 核心目标拆解:
- 按部门（DivisionID）统计总薪资（TotalDivisionSalary）；
- 找出每个部门薪资最高的员工姓名（Name）和薪资（TopSalary）；
- 计算最高薪资占部门总薪资的百分比（SalaryUtilization）；
- 判断是否存在预算优化潜力（最高薪资 < 部门总薪资 50% 则标 “Yes”，否则 “No”）。
- 技术方案:
- 先用
GROUP BY统计各部门总薪资； - 用窗口函数
ROW_NUMBER()按部门分组排序，筛选出每个部门薪资最高的员工； - 关联总薪资结果与最高薪资员工数据，计算百分比；
- Use
CASE WHENDetermine budget optimization potential.
- 先用
Python3 Challenge
Problem-solving ideas
- Supplementary core objectives (based on question context): The title does not fully list the final tasks, but combined with scenarios and examples, the core requirements are:
- Parse the flat array into structured review data (ratings, review content, date) by "grouping every 3 elements";
- Common derived requirements (complete reasonable tasks): statistical rating distribution, filtering comments on specified dates/ratings, extracting keywords, etc.
- 技术方案:
- Traverse the array in steps of 3 and parse comments in groups;
- Type conversion (score from string to integer);
- Encapsulate it as a function and return structured data (such as a list and nested dictionary) to facilitate subsequent analysis.
The hidden “killing point” of Amazon OA Coding
This is the most easily overlooked part by many question seekers.
Just run the sample case and submit it
Amazon OA's test case is quite "insidious":
- Empty input
- Limit length
- Repeating elements
- Special characters/boundary index
Can AC ≠ Can pass OA
No control complexity
Amazon values:
- Are you aware of the complexity issue?
- Whether to proactively optimize
Even if you don’t write the optimal solution,
But in the code or ideas, "I know where the bottleneck is", the results may be different.
Imbalance in time allocation
The real situation is:
- The first question took too long to write
- The second question was not finished at all
Amazon OA It is not necessary to get full marks on both questions,
But usuallyAt least one question must be "completely correct + high quality code".
Work Style Assessment (more important than you think)
Many people underestimate this part.
The essence of Work Style Assessment is to test:
- Do you meet the Amazon Leadership Principles
- Is your decision-making model "stable and consistent"?
LPs involved in high frequency include:
- Ownership
- Customer Obsession
- Bias for Action
- Dive Deep
A very common mistake is:
The logic behind the options is contradictory
Amazon is not afraid of your “imperfections”;
But I strongly reject “value swing”.
Leave the OA level to someone who is more familiar with the rules.
For OA from major manufacturers such as Amazon, Meta, Google, and Goldman Sachs, we provide Remote invisible OA assistance service, covering common inspection modules such as algorithms, SQL, Debug, and system questions. The entire process adopts a low-interference remote connection method. There is no need to cut the screen, install plug-ins, or trigger the platform risk control mechanism, ensuring the security of the account and test environment to the greatest extent.
All assistance is completed one-on-one by engineers with real experience in interviews and actual operations at major manufacturers. It is strictly implemented in accordance with OA time limits and scoring logic, with a focus on ensuring that the pass rate of hidden test cases and code quality are up to standard. We focus more on "whether we can advance to the next round" rather than just "finishing the question."
Applicable to mainstream OA platforms such as HackerRank, CodeSignal, Codility, and Niuke.com.
There is no charge for failed test cases, and the process is clear and traceable.
