Summer 2026 Google SWE Intern OA / Interviews: 42 Experiences (Part 2 :)

Continued from Part 1 of this three-part roundup. This section covers firsthand live-interview reports 13–28.

Coverage in this part: Google SWE/SDE Intern and PhD SWE Intern; live technical screens and virtual onsites; reports from October 2025 through January 2026.

Detailed candidate timelines, continued

13: Currency conversion with profitable-cycle handling

Report date: October 16, 2025 · Role: SWE Intern · Level: Bachelor's

In Maximum Currency Conversion With Arbitrage, the candidate received directed currency-rate strings and had to maximize the amount obtainable from a source currency to a destination through multiple hops. The follow-up was the important part: a profitable reachable cycle can make the answer unbounded, so a plain “visited” DFS is not enough.

14: Two optimized classics in New York

Report date: October 26, 2025 · Role: SWE Intern · Location: New York · Result: Offer

The candidate described two remote coding rounds and said optimized code plus time and space analysis were expected.

1. Russian Doll Envelopes, with width sorting and LIS on heights.
2. Next Greater Numerically Balanced Number.

Both problem identities were directly linked, so little reconstruction was needed.

15: Exact-digit clock, then an unspecified Dijkstra problem

Report date: October 27, 2025 · Interview date: September 2025 · Role: Software Engineer Intern · Result: No offer

The candidate had two 50-minute technical interviews and one behavioral question before coding. The first task, Next Time Using the Same Digits, asked for the next clock time using exactly the same four digits and gave 09:25 → 20:59. A second Dijkstra problem was mentioned but not specified enough to publish.

The stated clock example is internally awkward because 09:52 is an earlier exact-digit time. FastPrep follows the stated “next valid time” rule and calls out the conflict instead of silently copying the example.

16: Float-valued tree transformation and a tiny follow-up

Report date: October 29, 2025 · Role: Software Development Intern · Interview: 45 minutes

Transform and Prune a Mode-Valued Binary Tree combines several operations: replace each value by its reciprocal, swap left and right child pointers, find the mode, and remove subtrees rooted at mode-valued nodes. The interviewer added a division-by-zero check. The final follow-up became Swap Two Numbers.

17: Kadane with equal endpoint values

Report date: November 5, 2025 · Role: Software Engineer Intern · Result: Offer

The candidate reported two LeetCode-style interviews. One array question, Maximum Subarray Sum With Equal Endpoints, asked for a maximum-sum contiguous range whose endpoint values are equal and admitted a Kadane-like optimization. A second string-parsing task was mentioned but not defined.

18: Tree-dependent knapsack in a PhD loop

Report date: November 7, 2025 · Role: PhD SWE Intern · Result: Pass

Two 45-minute interviews happened in one morning. The preserved problem, Tree-Dependent Knapsack, uses a tree-shaped dependency graph: selecting an item requires selecting its parent. The report also illustrates interviewer variance—one interviewer was mostly quiet, while the other stayed engaged and moved through several tasks.

19: One exact array rule and two transparent reconstructions

Report date: November 8, 2025 · Role: PhD SWE Intern

The first problem asked for an array to be made “as sorted as possible” when each element may move right by at most k positions but may move left without limit. FastPrep formalizes that as Lexicographically Smallest Array With K-Limited Right Moves.

The source then preserved only two technique-level descriptions: streaming with a deque plus hash map, and a typical heap-based top-k question. Those became the clearly marked reconstructions First Unique Event in a Stream and Stable Top K Frequent Words; their exact interview contracts are not claimed.

20: Backtracking, graph traversal, and three manager calls

Report date: November 10, 2025 · Role: SWE Intern · Location: United States · Result: Declined offer

The process included coding, behavioral, two technical rounds, project matching, and conversations with three managers. The named questions were Subsets using backtracking and Number of Islands using graph traversal.

This is one of the clearer reminders that clearing the general technical loop and finding a host team were separate pieces of the process.

21: Cross-day logs under memory pressure

Report date: November 28, 2025 · Role: PhD SDE Intern · Result: Pass

In Cross-Day Object and Client Log Intersection, the candidate received two daily logs of (timestamp, objectId, clientId) and had to find object IDs present on both days and associated with at least two clients. Follow-ups asked for set-intersection complexity, then a memory-limited approach using sorting and two pointers.

22: An easy opener before Skyline

Report date: December 4, 2025 · Role: PhD SWE Intern · Stage: Additional interview

The candidate described a deliberately easy string-processing appetizer, followed by the substantive Skyline Problem. The opener's actual contract was not preserved; Run-Length Encode a String is explicitly a reconstruction. Skyline itself uses the standard sweep-line / heap formulation.

23: Recruiter call, OA, then Flood Fill

Report date: December 5, 2025 · Role: Software Engineer Intern · Result: Offer

The candidate reported a recruiter call, an online assessment, and two LeetCode-medium final questions. Flood Fill was named directly. The second final-round problem was not identified.

24: Longest duplicate substring and a shuffle follow-up

Report date: December 6, 2025 · Role: PhD SWE Intern

Round one directly used Longest Duplicate Substring. A later follow-up asked how to shuffle using random.randint(). The source did not preserve enough of the shuffle contract to turn it into a second standalone problem.

25: From course intervals to a calendar layout

Report date: December 10, 2025 · Role: Software Engineering Intern 2026 · Location: Shanghai · Result: Team matched

The author's two same-day interviews ran from 11:00–11:45 and 13:00–14:00, with shared-document coding, debugging, and iterative changes. The afternoon problem eventually asked for calendar rectangles, but the interviewer first separated Minimum Classrooms for Courses as a clean subtask.

Only that well-defined subtask became a standalone practice problem; the missing layout geometry was not invented.

26: Medium coding plus behavioral and hiring-manager rounds

Report date: December 16, 2025 · Role: SWE Intern · Result: Declined offer

The candidate described several rounds, including behavioral and hiring-manager conversations. The named coding task was Maximum Product Subarray, discussed interactively with the interviewer.

27: Tag matching and one-dimensional delivery

Report date: December 20, 2025 · Role: PhD SDE Intern · Result: No offer

1. Maximum Programmer-Problem Matching: match programmers to Stack Overflow-style questions by shared tags, starting with brute force and improving lookup with a hash map. A priority-queue follow-up was mentioned but not specified.
2. Maximum Tea Deliveries With Minimum Distance: place people and tea shops on a line, allow one delivery per person and shop within distance five, maximize the number delivered, then minimize distance among maximum-cardinality matchings.

The second task required recognizing a matching/greedy structure rather than reaching for a generic graph template.

28: Two Shanghai rounds with very different interviewers

Report date: January 7, 2026 · Role: SWE Intern · Level: Bachelor's · Location: Shanghai · Result: No offer

The first round switched from a brief Chinese greeting to English and asked for the Most Active Caller by Distinct Contacts, not message count. The second stayed in Chinese, included project questions, and became Batch Point-in-Range Queries: decide whether each query value—including possible floats—fell in any supplied range, followed by scale and complexity discussion.

The candidate's account is also a useful reminder to clarify the numeric domain instead of assuming integers.

Part 3 resumes at experience 29 and closes with the evidence-weighted practice priorities and final checklist.