Part 3 of the Google 2027 Intern interview tracker. This part covers reports 17–22 and closes with the practical takeaways.

Series: Part 1 · Part 2

Reported interviews 17–22

17: Duplicate trace IDs with an O(1)-space follow-up

Report date: Sep 15, 2026 · Role: SDE Intern · Stage: VO

Round 1

• Detect Duplicate Trace IDs in a Linked List: each node stores a string trace ID. Return whether any ID appears more than once.
• The base solution walks the list and stores seen IDs in a hash set.
• The follow-up forbids extra space. Before answering, clarify whether the list may be modified, whether the output is still boolean, and whether trace IDs can be ordered or compared beyond equality.

Round 2

• The source only preserved the labels “BFS + min-heap.” There was not enough information to recover the actual task.

18: Duplicate number and matrix zeroing

Report date: Sep 14, 2026 · Role: SDE Intern · Stage: VO

Question 1

• Find the Duplicate Number: an array contains n + 1 integers and exactly one distinct duplicate. Use O(1) extra space and beat O(n²) time.
• The follow-up removed the “one duplicate” or “values from 1 through n” assumption. Floyd cycle detection cannot be reused without first rebuilding the contract.

Question 2

• Set Matrix Zeroes: if a cell is zero, set its full row and column to zero, in place.
• The follow-up made the matrix too large to fit in memory and asked about blocks, external storage, and multiple passes.

19: Prefix remainders with three useful follow-ups

Report date: Sep 11, 2026 · Role: SDE Intern · Stage: Phone screen

• Shortest Subarray With a Target Remainder: the base task asked whether a nonempty contiguous subarray had a sum whose remainder modulo 70001 was k.
• Follow-up 1 added negative values. Since languages handle % differently, normalize every remainder into [0, 70000].
• Follow-up 2 changed the goal to the shortest length. Store the latest matching prefix index, not only the earliest one.
• Follow-up 3 used roughly one million streamed values under a memory limit. Since the modulus is fixed, a 70001-entry array can replace a hash map while the program keeps only a rolling prefix remainder.

20: Count how many people each person can see to the left

Report date: Sep 5, 2026 · Role: Software Engineer Intern · Stage: VO

• Count Visible People to the Left: people have distinct heights. A person can see someone to the left when nobody between them is taller than both endpoints.
• In the warm-up array [1, 10, 6, 7, 9, 8, 2, 4, 3, 5], the last person can see 6 people.
• The follow-up asks for the count for every person.
• Scan left to right with a monotonic stack. Each value is pushed and popped at most once, for O(n) time and O(n) space.

21: Minimum flight hops and the median of the latest K response times

Report date: Aug 31, 2026 · Role: SWE Intern · Stage: VO · Source note: The Aug 30 and Aug 31 posts describe the same rounds and follow-ups, so they are merged as one cross-posted interview

Round 1

• Cities and bidirectional flights form an unweighted graph. Find the minimum number of hops from the start to the destination. FastPrep maps the same core task to Shortest Hop Path Between Machines.
• BFS solves the base problem. Follow-ups ask for the full route, disabled flights, and a full time-and-space analysis.

Round 2

• Sliding Window Median of Response Times: continuously receive response times and return the median of the latest K values.
• Use a max-heap for the smaller half and a min-heap for the larger half.
• The hard part is deleting expired values when the window moves. Other follow-ups cover high data volume, lower memory use, duplicates, and even-sized windows.

22: Return a book through a grid, then Top K stars

Report date: Aug 5, 2026 · Role: Software Engineer Intern · Stage: VO · Format: Two rounds

Round 1

• Minimum Distance to Return a Book: an m × n grid contains a start cell, libraries, and train stations. A station can reach a library with a known extra distance.
• Run BFS from the start. Compare walking directly to a library with walking to a station and then adding that station's library distance.
• One sentence in the source said “minimum walking distance,” while the given formula added the station distance. The practice page uses the formula because it is the more concrete, testable rule.

Round 2

• The second task was again K Closest Stars from a Data Stream.
• The candidate compared sorting and a min-heap before choosing a max-heap of size K for O(N log K) time.
• Tests included K = 0, K > N, and equal distances.

What to prioritize

1. BFS and DFS, including path recovery.
2. Prefix sums and difference arrays, including inclusive range boundaries.
3. Two pointers, sliding windows, and clean binary-search boundaries.
4. The right heap: max-heap for Top K smallest, min-heap for the earliest free resource, and two heaps for a median.
5. State and identity: know when visited[node] is enough, when a remaining budget belongs in the state, and whether duplicate values can identify a node.
6. Streaming, batching, memory limits, concurrency, timeouts, and failure rules.