Many people are there Microsoft You will get stuck in OA, but looking back it is not how difficult the question itself is, but some very typical problems that appear repeatedly: for example, the direction is right, but it becomes messy as soon as you write it; for example, you have an idea, but you always hesitate at the key turning point; or you subconsciously use simulation to hard-top, and the complexity explodes. Behind these situations, they actually point to the same problem: high-frequency question types have not really settled into a stable and reusable model. Today I will share a very typical Microsoft OA example.
Interview Process Timeline (Reference)
This round of process takes a long time In summary, and there is an obvious waiting period in the middle. It belongs to Microsoft's usual rhythm, so there is no need to be overly anxious.
- OA received approximately 1–2 weeks after delivery
- 10/3 Received OA
- 10/16 Receive final round notification
- 10/26 Final round (three rounds in total)
- 11/8 Recruiter email description large volume of recruiting
- 11/21 Action Center status changes to complete
- 12/11 Formal offer
The overall pace is not fast, and the waiting time is long, but it is Microsoft's usual style.
Q1: Maximize the MEX of the array (sorting + greedy trap model)
Core of the question: Given an array of integers, you can operate on any element and reduce it to any integer in the interval [0, original value]. The goal is to maximize the MEX (minimum missing non-negative integer) of the entire array by adjusting the array.
If this question is written from the perspective of "construction" or "step-by-step simulation", it is easy to become more and more confusing. The real key is to first think clearly about the essential constraints of MEX.
If you want MEX = m, you must also satisfy:
- There are at least 0,1,2,…,m-1 in the array
- Each number can only be reduced, not increased
- Large numbers can be used to fill small missing values, but decimals cannot be used if they are already too small.
Once you understand these points, the model is very clear: sorting + greedy trap.
If you want to make MEX larger, you essentially want to see if you can start from 0 and "occupy" consecutive non-negative integers one by one. Once you realize that each number can only be changed to a small size, and large numbers can be used to make up for the pitfalls of decimals, your ideas will naturally converge. After sorting the array, start from the minimum required value 0 and scan along. As long as the current element is not less than the number you are missing now, reduce it to the exact required value to occupy this position; if the current element is already smaller than the required value, it means that it has lost the ability to make up for the hole, so just skip it. The entire process requires only one linear scan, no backtracking, and no complex states. Wherever the data can be filled continuously all the way, that position is the maximum MEX that can be achieved. What really differentiates people in this question is whether they choose the right model at the beginning, rather than how fast they write the code.
Q2: String Roll operation (differential array/prefix sum optimization)
Title description: Given a string s and an array roll, each roll[i] represents an alphabetical loop +1 (a→b,…,z→a) for the first roll[i] characters of s, performs all roll operations in sequence, and outputs the final string.
The intuitive way to write it is to modify the prefix characters every time you roll, but in the worst case the complexity will degrade to O(n²), and OA will basically require TLE.
If we start with each roll operation itself, it is easy to subconsciously simulate prefix modifications, but this path is unfeasible in terms of complexity. Think about it from another perspective, each character ultimately only cares about one thing: how many times it has been added. All roll operations are essentially interval additions to string prefixes. These additions are first accumulated using differential arrays, and then restored to each position through a prefix sum. Finally, the alphabet circular mapping is performed uniformly. This modeling method of switching from "process" to "result" is the core of this question and is also very consistent with Microsoft's preferred engineering thinking in OA.
OA review summary
If you have a feeling when reading these questions: the questions are familiar and the ideas are roughly known, but when it comes to actually writing code, you always feel that it is not stable enough, and you even lose points inexplicably in OA, then the problem is often not in the number of questions. The more common reason is that high-frequency question types still remain at the level of "on-the-spot thinking" and have not truly settled into a set of stable and reusable thinking models. Microsoft's question style is just perfect for testing this.
Why more and more students are choosing interview assistance
In actual contacts, we found that many of the students who came for consultation were not incompetent at the basics. Instead, they had already passed a lot of questions and roughly knew what the test was about. However, their performance was unstable in OA or formal interviews: model judgment was slow, implementation details were wrong, or they hesitated at key nodes, and ultimately missed the offer. Because of this, more and more students are actively looking for interview assistance, hoping to steadily use their inherent abilities in high-pressure, time-limited real-world scenarios.
In the past period of time, we have assisted many candidates to successfully go through OA, VO to the final round, and finally got their favorite offer. If you are also in the stage of preparing for OA or interview and want to avoid detours and improve your pass rate at critical moments, you can directly Contact us .
