As of October 2026, public candidate reports for Google Summer 2027 SWE Intern (U.S.) still center on two 45-minute live coding rounds focused on algorithms and data structures. Candidates clarify the prompt, explain their approach, analyze complexity, write code, dry-run it, and handle follow-ups.

Google is also piloting a Gemini-assisted code-comprehension interview with some U.S. teams. However, there is no public evidence that this format is broadly used for Summer 2027 Intern interviews.

Of course, always go with whatever your recruiter tells you and follow the format in your interview invitation.

Coverage: Google SWE / SDE Intern phone screens and virtual onsites reported from August 5 to September 30, 2026. Went through all the reports I could find so far, merged one duplicate from the same interview, and ended up with 20 unique interview records. As far as I can tell, they’re all U.S. based.

Hope this series helps all my lovely frenz! 💛

I’ll keep updating this series throughout the recruiting season.

The 2027 Intern Interview Era Begins ദ്ദി(˵ •̀ ᴗ - ˵ ) ✧

The first solution was often straightforward. The follow-ups made the interview hard.

1. Graphs, trees, and grids dominated. Thirteen records used graph, tree, or grid search.
2. A working answer was only step one. Follow-ups added path output, broken edges, streams, memory limits, dynamic data, or thread safety.
3. Edge cases came early. Examples included K = 0, touching intervals, duplicates, even windows, negative remainders, and unreachable states.
4. Several questions became production discussions. Candidates had to talk about batching, timeouts, metrics, memory protection, and tests.

Prep order: drill BFS, DFS, binary search, two pointers, sliding windows, heaps, and topological sorting. Then dry-run the code and ask what changes for streaming or limited memory.

22 reported interviews, newest first ᕙ( •̀ ᗜ •́ )ᕗ

01: An optimal first round, then a rule-interpretation mismatch

Report date: Oct 2, 2026 · Role: Summer 2027 SWE Intern · Location: United States · Status: Awaiting an outcome

Round 1

• The candidate spent roughly 25 minutes clarifying the prompt, explaining the approach and complexity, coding, and dry-running the result.
• The interviewer confirmed that the solution was optimal.
• Two follow-ups covered the brute-force complexity and an extension of the original problem. The candidate discussed the extension and its edge cases without coding it.

Round 2

• The first approach had a flaw that the interviewer exposed with a counterexample. The candidate changed direction quickly and reached a linear-time approach that the interviewer approved before implementation.
• During the dry run, the candidate and interviewer expected different results because they had interpreted one rule differently.
• Around 35 minutes in, the interviewer added a harder generalization. The candidate identified the intended technique and wrote the full algorithm and complexity, but time expired before one remaining concern was resolved.

Interview format

• There was no dedicated behavioral section beyond a short introduction and closing questions.
• The candidate said the rounds were technical, while communication and Googleyness were still evaluated through the problem-solving discussion.

The candidate did not disclose either prompt, so there is no honest question-specific FastPrep link for this report.

02: 0-1 BFS, then negative edges

Report date: Sep 30, 2026 · Role: Software Engineer Intern · Stage: VO · Result: Strong hire, according to the post title

Round 1

• Minimum Direction Violations: The candidate received a directed graph, but every edge could be used in either direction. Following the listed direction cost 0; going against it cost 1. The task was to find the minimum cost from start to end.
• The expected solution was 0-1 BFS. Put a zero-cost move at the front of a deque and a one-cost move at the back. The total time is O(V + E).
• The first follow-up allowed negative edge weights. At that point 0-1 BFS no longer works; the report moved to Bellman-Ford.
• The next follow-up added multiple sources and destinations. The suggested reduction used a super-source and a super-sink.

Round 2

• Count Triplets With Sum at Most Target: count index triplets whose sum is no greater than a target.
• Sort the array, fix the first value, and use two pointers on the remaining suffix. When the current sum is small enough, a whole range of partners can be counted at once. The final time is O(n²).

03: Word Break without repeated work

Report date: Sep 30, 2026 · Role: Software Engineer Intern · Stage: VO

• Word Break: given a string s and a dictionary, decide whether the whole string can be split into dictionary words.
• The report used dfs(i) for the suffix beginning at index i. Try each possible next word, then recurse from the next index.
• Convert the dictionary to a set and memoize each dfs(i) result, since different splits can reach the same suffix.
• The reported complexity was O(n²) time and O(n) extra space.

04: Interval conflicts and rental-car reuse

Report date: Sep 30, 2026 · Stage: Phone screen

Question 1

• Target Interval Conflict: given one target interval and a list of intervals, decide whether any interval conflicts with the target.
• Intervals are half-open: [start, end). Two intervals do not overlap when a.end <= b.start or b.end <= a.start.
• For one query, a linear scan is enough. Sorting only starts to make sense when there are many queries or the caller needs every conflict.

Question 2

• Minimum Cars for Rental Requests: Each rental request had an order ID, pickup time, and return time. Assign requests to cars so that no two requests on one car overlap, then return each car's full request list.
• Sort by pickup time and keep cars in a min-heap by next available time. Reuse the heap top when possible; otherwise create a car.
• With only K cars, report failure or the unassigned request when every car is busy. Do not overwrite an assignment.
• The post called out touching endpoints, equal start and end times, unsorted input, and K = 0.

05: Took it down because I wasn’t fully sure it was real...

06: Top K, a concurrent scheduler, and an N-ary tree codec

Report date: Sep 30, 2026 · Role: SDE Intern · Stage: VO · Format: Two rounds, about 45 minutes each

Round 1

• K Closest Stars from a Data Stream: process a huge stream of (star_id, distance) pairs and return the K smallest distances.
• Sorting everything costs O(N log N). A max-heap of size K brings the time down to O(N log K).
• The interviewer asked about K = 0, K > N, and equal distances.

Round 2

• Concurrent Priority Task Scheduler with Timeout Cancellation: support priorities, timeout cancellation, and thread safety. The reported direction used a priority queue and a ReentrantLock.
• N-Ary Tree BFS Serialization and Deserialization: serialize and deserialize an N-ary tree with BFS, including null markers and level boundaries.
• The hard part was not naming the data structure. It was finishing the concurrency rules and encoding format under time pressure.

07: 2D prefix sums, then an unfamiliar difference array

Report date: Sep 30, 2026, based on the screenshot date · Role: Summer 2027 SWE Intern · Stage: Two back-to-back coding rounds · Result: Rejected

Round 1

• The candidate described the first question only as a 2D prefix-sum problem.
• The screenshot did not show the matrix operation, input, output, constraints, or follow-ups, so I am keeping this as a topic instead of inventing a question.

Round 2

• The candidate identified the question as Zero Array Transformation I.
• Each query covers an inclusive range. For every covered position, you may choose whether to reduce that value by one. The task is to decide whether all values can become zero after all queries.
• A difference array counts how many queries cover each index. After one prefix scan, every index needs coverage of at least its original value.
• The candidate had never practiced difference arrays, spent much of the round stuck, and only managed to write an answer near the end. They later said the problem was lack of fluency, not that the task was impossible.

Interview format

• Both rounds were algorithmic coding interviews and ran back to back.
• The candidate said there was no BQ.

08: Full follow-ups first, implementation-heavy OOD second

Report date: Sep 30, 2026 · Role: SWE Intern · Location: United States · Status: Awaiting an outcome

Round 1

• The candidate needed a small amount of guidance and help correcting one line of code.
• They recovered, finished the rest of the task, completed every follow-up, and reached the optimal solution.

Round 2

• The interviewer prompted the candidate toward the optimal data structure.
• The problem was implementation-heavy and object-oriented. The candidate estimated that they completed roughly 70–80% of the code.
• The overall design and logic were sound, but several helper or class methods still contained implementation mistakes.
• Near the end, the interviewer asked the candidate to leave those helpers and implement one key function directly. They completed that function and then answered the time and space complexity questions.

The source did not disclose the object model or required operations, so there is no exact problem match. For transferable implementation-heavy practice, not as a claim about the withheld interview prompt,can probably try Concurrent Priority Task Scheduler with Timeout Cancellation.

09: A request window and a shortest path through machines

Report date: Sep 29, 2026 · Role: SDE Intern · Stage: VO

Round 1

• High-Traffic IPs in a Sliding Window: logs arrive in time order. Find every IP with more than 100 requests inside any ten-minute window.
• The candidate started with brute force, then moved to a sliding window plus a hash map for a linear scan.
• Follow-ups covered very large data, streaming input, lower memory use, and missed boundary cases.

Round 2

• Shortest Hop Path Between Machines: find the fewest hops from machine A to machine B and print the path.
• BFS solves the base problem. Returning the path requires a parent map.
• Follow-ups covered huge graphs, cycles, full path recovery, and lower extra space.

Continue with Part 2 · Part 3.

And here’s a group photo of everyone in the Secret Base 🫶

Sources: Pretty much anywhere online with useful interview info. AKA: trust me bro ૮ • ﻌ - ა