Articles
Tower Research SWE OA 2027: Camera Proctoring, MEX Problems, and Difficulty
Prepare for the Tower Research SWE OA 2027: dual-camera reports, MEX and DP strategy, difficulty, and a focused practice plan.
A June 29, 2026 candidate report made the Tower Research SWE OA 2027 look unusually intimidating: front and back cameras were reportedly on, and the remembered coding task was not a familiar two-pointer exercise. It asked candidates to count permutations according to the distinct MEX values produced by all contiguous subarrays.
That is a valuable recruiting-season signal, but it is not a universal policy document. This guide separates the current India campus report from older Tower formats, explains the MEX reasoning, and gives you a preparation plan that still works if your batch receives different questions.
The goal is not to memorize one leaked prompt. It is to become fluent in the kind of structural reasoning that a hard OA can test.
Tower Research SWE OA 2027 preparation: current dual-camera reports, MEX reasoning, and difficulty.
Quick Answer: What Should You Expect?
The safest conclusion is to prepare for strict proctoring and contest-style algorithmic reasoning, while treating your invitation as the only authority on the exact setup. One current candidate described front and back cameras and reconstructed a hard MEX counting problem. The post did not confirm the platform, timer, total question count, cutoff, or whether every 2027 batch uses the same form.
| Question | Best current answer | Confidence |
|---|---|---|
| Is the Tower Research OA camera-proctored? | One June 2026 India campus report says both front and back cameras were on. | Current candidate report, not universal policy |
| What problem was reported? | A permutation-counting problem based on the distinct MEX values of all contiguous subarrays. | Reconstructed from memory |
| How difficult is it? | Medium-hard to hard for a LeetCode-only candidate; more approachable with contest combinatorics and DP practice. | Reasoned assessment, not an official rating |
| Is the platform HackerRank? | HackerRank appears repeatedly in prior Tower India reports, but the current post did not name its platform. | Historical pattern only |
| What else should you study? | DP, graphs, trees, debugging, complexity, OS, networking, DBMS, and language fundamentals. | Repeated across earlier SWE reports |
What Is Confirmed for the 2027 Tower Research OA?
The current signal comes from one detailed Reddit post about a 2027 graduate SWE assessment. The author said a front camera and a back camera were active, so they reconstructed the question afterward rather than photographing it. The remembered constraints included up to 1,000 test cases, N up to 5,000, and a total N of at most 5,000.
Comments on that post disagreed about difficulty. One reader compared it with a roughly 1400-level competitive-programming task; others struggled to derive the recurrence. A later commenter linked CodeChef START212A P6BAR as the same or a closely matching problem. That link makes a useful practice reference, but it does not prove Tower will repeat it.
Older reports show why a single format should not be overgeneralized. A 2025 on-campus candidate described a two-hour HackerRank test with 10 core-CS MCQs and three coding questions in dynamic programming, graphs, and trees. Another intern report described three hard coding problems, a C++ debugging task, and MCQs. A backend SDE-1 candidate instead reported 135 minutes, two coding tasks, and four theory questions.
Use current reports to guide preparation, but use your own invitation for operational rules.
How Tower Research Camera Proctoring May Work
“Front and back camera” usually implies two views of the testing environment, often a computer camera plus a second phone or device. The current report supports that possibility for one India campus batch. It does not establish which vendor Tower used, whether the feeds were stored, or whether a human watched in real time.
Does Tower Research record your screen?
The public evidence does not answer that. Camera access, screen capture, browser monitoring, and post-test review are separate controls. A camera being on does not automatically prove the screen was recorded, and an older HackerRank report does not prove the 2027 form uses the same settings.
Read every instruction before starting. Look for permissions involving webcam, microphone, screen sharing, browser extensions, a secondary-device QR code, identity verification, tab switching, and copy-paste. If the invitation conflicts with anything online, the invitation wins.
Do a legitimate hardware rehearsal
Test the exact computer, browser, camera, charger, and network you plan to use. Place the second device on a stable stand, verify that your face and workspace are visible, close unrelated applications, and clear the desk. This is not about evading monitoring; it prevents a technical failure from obscuring a strong performance.
The Reported MEX Problem, Explained
MEX means the minimum excluded value: the smallest non-negative integer missing from a set. For example, the MEX of [0, 2, 3] is 1. The reported problem considered a permutation of 0...N-1, collected the MEX of every non-empty contiguous subarray, and asked how many permutations produce exactly K distinct MEX values.
A brute-force interpretation is hopeless. There are N! permutations and O(N²) subarrays per permutation. The constraints signal that the intended solution counts whole classes of permutations without generating them.
The structural observation
For an intermediate value m to be the MEX of some subarray, that subarray must contain every value from 0 through m-1 and exclude m. Therefore, m is reachable when the position of m lies outside the smallest interval containing the positions of all smaller values.
That converts a question about all subarrays into a question about how the positions of increasingly large values expand an interval. It is the key step: once you see the interval property, you can count permutations by inserting values in increasing order.
Why insertion creates a DP recurrence
Suppose you already arranged 0...n-2 and insert n-1. There are n insertion slots. Putting the new largest value at either end preserves the previous full interval and creates one additional reachable MEX; placing it in one of the n-2 interior slots replaces the previous maximum MEX without increasing the number of distinct values.
That gives the common transition shape, for n ≥ 3:
dp[n][k] = 2 * dp[n-1][k-1] + (n-2) * dp[n-1][k]
You must handle the tiny base cases explicitly, because MEX 0 appears only once a non-empty subarray can exclude the value 0. Enumerate every permutation for n = 1, 2, 3 before trusting the recurrence. The resulting DP is O(NK) time and can use rolling rows for O(K) memory.
The reported MEX task becomes manageable after proving the interval and insertion properties.
How Difficult Is the Tower Research SWE OA?
For a candidate who only recognizes standard LeetCode templates, the reported problem is legitimately hard. It combines a definition that looks local, a global set over every subarray, permutation counting, a structural proof, dynamic programming, and modular arithmetic. The implementation is short only after the proof is correct.
A competitive programmer may rate the same task much lower because insertion recurrences are familiar contest machinery. That explains the conflicting comments better than arguing over one unofficial rating. The difficulty gap is not typing speed; it is whether you can discover and justify the invariant under time pressure.
| Skill | What the reported problem demands | How to practice |
|---|---|---|
| Modeling | Replace all-subarray enumeration with a positional condition. | Practice turning range statements into prefix, span, or contribution arguments. |
| Proof | Show exactly when a MEX value is reachable. | State necessary and sufficient conditions before coding. |
| Counting DP | Count two end placements versus interior placements. | Solve insertion, combinatorics, and “build the permutation” DP tasks. |
| Validation | Catch small-case and MEX-0 edge conditions. | Brute-force tiny inputs and compare them with the DP. |
What Else Should You Prepare?
Do not spend your entire week memorizing one recurrence. Historical Tower SWE assessments repeatedly touch dynamic programming, graphs, trees, debugging, time complexity, operating systems, networking, databases, OOP, and language behavior. The exact mix changes by role and campus.
For coding, prioritize problems that require a proof rather than a named template: tree DP, graph invariants, constructive algorithms, prefix/suffix preprocessing, combinatorial counting, and optimized state transitions. Add two timed contests so you practice deciding when to abandon a dead approach.
For fundamentals, be able to explain process versus thread, virtual memory, deadlocks, TCP behavior, indexing, transactions, inheritance, polymorphism, and the complexity of common data structures. If interviews follow the OA, broaden into system design questions and prepare for performance-oriented follow-ups consistent with Tower's official description of core engineering work.
A Focused 7-Day Preparation Plan
| Day | Work | Deliverable |
|---|---|---|
| 1 | Rebuild MEX basics, permutation positions, and tiny-case brute force. | One page of examples for n = 1 through 4 |
| 2 | Solve three counting-DP problems with insertion or contribution arguments. | For each, write the state and proof before code |
| 3 | Practice one graph, one tree, and one prefix/suffix problem under time. | Three accepted solutions with complexity notes |
| 4 | Debug a broken C++ or Java program and review language fundamentals. | A bug log: symptom, hypothesis, evidence, fix |
| 5 | Review OS, networking, DBMS, OOP, and time-complexity questions. | Twenty concise spoken explanations |
| 6 | Run a full timed contest-style mock with strict rules and custom tests. | Post-mock analysis of wasted time and missed edges |
| 7 | Rehearse the camera setup, retake weak topics, and stop early enough to rest. | A clean test environment and final checklist |
Test-Day Strategy for a Hard OA
Spend the first minutes classifying, not coding
Read every task and its constraints before committing. Identify whether each problem is implementation-heavy, proof-heavy, or likely to need a known data structure. A clean solution to the most tractable problem is usually more valuable than forty minutes of untested code for the hardest one.
Write the invariant in plain English
For a problem like MEX counting, write one sentence that must be true: “MEX m is reachable exactly when m lies outside the span of smaller values.” Then attack that sentence with counterexamples. Coding before the invariant is stable produces fast wrong answers.
Use tiny brute force as an oracle
If the optimized solution is a recurrence, write or mentally enumerate tiny cases. Compare expected counts, check modulo placement, and test the smallest legal inputs. Hidden tests often expose initialization and boundary errors rather than the main idea.
What Happens After the Tower Research OA?
There is no verified universal cutoff or response timeline for this 2027 form. One 2025 on-campus account received a shortlist after two days and said both OA performance and resumes mattered. Other roles and recruiting channels can move differently, so a perfect-looking score should not be treated as a guaranteed interview.
Prepare the next stage while you wait. Review one project deeply, practice explaining design and performance trade-offs, and prepare concise behavioral and leadership stories about ownership, debugging, disagreement, and learning. Tower's official careers page emphasizes technical skill, clear communication, calm focus, and entrepreneurial thinking.
Frequently Asked Questions
Does the Tower Research OA require a camera?
One June 2026 India campus report for the 2027 graduate SWE process says both front and back cameras were active. That is strong evidence for that batch, not proof of a worldwide rule. Check your invitation and system test for the cameras, devices, permissions, and room view required in your assessment.
Does Tower Research record your screen or tab switches?
The current public report does not establish screen recording or tab-switch policy. Camera feeds, screen capture, browser monitoring, and copy-paste controls are separate settings. Read the test rules shown before launch and follow them exactly; do not infer permission from an older candidate's platform or experience.
Is the Tower Research OA on HackerRank?
Several historical Tower India SWE reports name HackerRank, including 2025 accounts. The June 2026 MEX post did not identify its platform, so HackerRank is a useful historical pattern rather than a confirmed 2027 fact. Your assessment link and technical check will provide the definitive answer.
Will Tower ask the same MEX problem again?
Do not count on it. The remembered problem is useful because it reveals a style: structural observations, combinatorial counting, and dynamic programming under tight constraints. Practice the recurrence, then solve adjacent MEX, insertion-DP, and constructive problems so you can transfer the reasoning when the surface story changes.
Is LeetCode enough for Tower Research?
LeetCode builds core DSA fluency, but the current problem rewards contest-style discovery and proof. Add timed contests, counting DP, constructive algorithms, and small-case brute-force validation. Historical reports also justify reviewing debugging and core CS subjects instead of preparing only algorithm templates.
Prepare for the Signal, Not the Rumor
The useful signal is clear: Tower Research can use demanding algorithmic assessments, and one current 2027 India campus report describes dual-camera proctoring plus a non-trivial MEX counting problem. The rumor would be assuming every candidate gets that exact setup, question, or cutoff.
Practice under a timer, prove the transition before typing, and compare your answer with a full solution. That preparation survives a changed story, a different batch, and a harder follow-up.