Google 2027 Intern Interview Experiences & Question Tracker (Part 2 :)

Part 2 of the Google 2027 Intern interview tracker. This part covers reports 10–16.

Continued from Part 1.

Reported interviews 10–16

10: File wildcards and a grid with removable obstacles

Report date: Sep 29, 2026 · Role: Software Engineer Intern · Stage: VO · Result: The post said the candidate received a pass email

Round 1

• Match File Paths with Wildcards: a single * matches any string inside one directory level, including the empty string.
• The follow-up added **, which can cross any number of directories. /home/**/a.txt can match both /home/a.txt and /home/user/docs/a.txt.
• The interviewer asked for a manual dry run and cases such as consecutive wildcards and equal filenames in different directories.

Round 2

• Shortest Grid Path with Obstacle Elimination: move from the top-left to the bottom-right while removing at most K obstacles.
• A BFS state needs the row, column, and remaining obstacle budget. (row, col) alone is not enough.
• The memory follow-up asked how to avoid a large 3D visited table. If the same cell was reached earlier with a better remaining budget, the weaker state could be pruned.
• The last discussion covered validation, timeouts, memory protection, metrics, and extreme-input tests.

11: Duplicate values made node identity the real problem

Report date: Sep 29, 2026, based on the screenshot date · Role: Summer 2027 SWE Intern · Stage: Two back-to-back technical rounds · Result: Unknown

Timeline

• Campus outreach arrived in late August or early September.
• The candidate completed the OA on September 14 and had two back-to-back interviews on September 28.

Question clue

• One question used a Node class whose visible data was an integer, even though different nodes could have the same value. The candidate needed a way to tell which node was which.
• They got stuck on the representation. Later, they thought about a separate id() or an index such as self.i assigned during traversal.
• The screenshots never reveal whether this was a tree, linked list, graph, or class-design task. They also omit the required operation and return value, so I am not turning it into a made-up problem.

Interview notes

• The candidate completed one of the two questions and did not finish the other.
• Both interviewers were friendly and offered guidance, but 45 minutes still felt tight once explanation and coding were included.
• Their main takeaway was to clarify assumptions early. An interviewer may guide you without directly pointing out the first wrong assumption.
• At the time of the post, the candidate did not know whether Google would add another round.

12: Count the top node of each island in a binary tree

Report date: Sep 26, 2026 · Role: Software Engineer Intern · Stage: Phone screen

• Count One-Valued Regions in a Binary Tree: every node is 0 or 1, and connected 1-valued nodes form a region. Return the number of regions.
• There is no need to collect every node in each region. Every region has exactly one highest node: its value is 1, and its parent is missing or has value 0.
• Run DFS and pass the parent along with the current node. Count each node that meets that rule.
• The reported complexity was O(n) time and O(h) recursion space.

13: Three Sum and level-order traversal

Report date: Sep 22, 2026 · Role: SDE Intern · Stage: VO

Round 1

• Three Sum: Unique Zero-Sum Triplets: sort the array, fix one value, and move two pointers inward.
• The follow-up asked how to handle many duplicates. The fixed value and both moving pointers all need duplicate skipping.

Round 2

• Binary Tree Level Order Traversal: use a queue and record the current queue length before processing each level.
• The follow-up asked for a DFS version. Pass the depth into the recursion and append to the matching result level.

14: Half-open interval subtraction, then an incomplete graph clue

Report date: Sep 21, 2026 · Role: SDE Summer Intern · Stage: VO

Round 1

• Subtract One Half-Open Interval from Another: return the part of interval A that is not covered by interval B.
• For A = [2.5, 7.5) and B = [4.3, 9.3), the answer is [2.5, 4.3).
• The result can be empty, one interval, or two intervals. [1, 3) and [3, 5) only touch, so they do not overlap. Empty output pieces such as [3, 3) must be removed.

Round 2

• The post only said the question used graph BFS and a min-heap. It did not include the state, input, output, or goal.
• That is not enough to rebuild a real problem, so I am leaving it incomplete.

15: Where each water drop ends, then the longest word from nine letters

Report date: Sep 21, 2026 · Role: Software Engineering Intern · Stage: VO

Round 1

• Trace Every Water Drop to Its Resting Destination: start one drop on every cell of a height grid. Each drop follows the steepest downhill move until it reaches a basin or leaves the grid.
• The report compared plain DFS, memoized DFS, and processing cells from low height to high height. The key was reusing the destination already found for the next downhill cell.
• The follow-up asked about memory pressure on a very large grid.

Round 2

• Longest Dictionary Word from Nine Letters: given a dictionary and nine available letters, return the longest word that can be formed without overusing a letter.
• Precompute a frequency table for every dictionary word, then check candidates from longest to shortest.
• The follow-up compared one query with many queries against the same dictionary.

16: Two binary searches, then course ordering

Report date: Sep 17, 2026 · Role: Software Engineering Intern · Stage: VO · Format: Two 45-minute rounds

Round 1

• First and Last Target Position in a Mountain Array: the array rises and then falls and may contain repeated values. Return the first and last target index.
• Start by explaining the linear scan, then move to O(log n). Use separate binary searches for the left and right boundaries.
• Follow-ups asked about an unsorted array, growing data, every matching interval, and boundary cases.

Round 2

• Find a Valid Course Completion Order: given courses and prerequisites, decide whether all courses can be completed and return any valid order.
• Build the graph, count indegrees, and run Kahn's BFS topological sort.
• Follow-ups asked for an actual cycle, work on only one subgraph, full dependency paths, and lower memory use.

Continued in Part 3.