Google Software Engineer Onsite Questions: 10 Experiences
- Company
- Reports
- 10 interview reports
- Level
- Unknown
- Stages
- Onsite
Google software engineer onsite questions reported in October 2026 by 10 candidates: 4 interns, 1 new grad, 2 experienced hires and 3 with no cohort given.
At a glance
- Interns (01–04): 2 of 4 described two coding rounds (02, 03). Graph tasks came up in 3 (01, 03, 04); no task was named twice.
- New grad (05): a review-summary pipeline design and a timed-trip coding task.
- Experienced (06–07): 06 got stuck on a digit-counting follow-up. 07 had a graph task, CPU scheduling and an auction class they called very simple, then was levelled from L4 to L3.
- Cohort not given (08–10): a fuel-route search they felt went badly (08), a hardware-timer task with three Googleyness questions (09), a monitoring design (10).
- Practise first: Minimum Union of Two Routes (intern 04), Find the Town Judge (intern 03), Finish Time of Fixed-Length CPU Tasks (experienced 07).
Coverage: Posted October 3–29, 2026; dated interviews run from April–June 2026 (07) to October 2026 (06). Interns are in the 2027 summer cycle where a year was given. No count below mixes levels.
Internship: Software Engineer Intern
What came up
Of 4 intern reports, 3 had graph work (01, 03, 04), 2 array or string tasks (01, 02) and 1 a tree task (02).
Questions we found
Dates are report dates.
| Question | Reports | Last seen |
|---|---|---|
| Course Schedule (changed version) | 1 | Oct 2026 |
| Non-Decreasing Array | 1 | Oct 2026 |
| Longest Substring Without Repeating Characters | 1 | Oct 2026 |
| Binary Tree Right Side View | 1 | Oct 2026 |
| Five-Minute Hit Counter | 1 | Oct 2026 |
| Find the Town Judge | 1 | Oct 2026 |
| Minimum Union of Two Routes | 1 | Oct 2026 |
Timelines
01: A changed course schedule and a non-decreasing array
Report date: Oct 16, 2026 · Role: SWE Intern, 2027 summer
A coach posted this for a student and named the tasks only: Course Schedule with an undescribed change (Course Schedule Cycle Detection is the base), then a non-decreasing array task (Make an Array Non-Decreasing With One Modification is our one-change version).
02: Longest substring and right side view, no way to run code
Report date: Oct 8, 2026 · Role: SDE Intern
Two 45-minute rounds on one day with a short break, pure algorithms: no resume deep dive, no behavioral questions. The editor couldn't run code, so the candidate traced each solution by hand and then gave its time and space cost.
- Round 1: Longest Substring Without Repeating Characters ("abcabcbb" gives 3). A sliding window with each character's last position; the left edge never moves back. Follow-up: return the substring itself by saving the start whenever the best length grows.
- Round 2: Binary Tree Right Side View, by level order (last node per level) or a right-first depth search (first node per depth). Follow-up: group the rightmost nodes by level. Five minutes were left for questions.
03: Five-minute hit counter and the town judge
Report date: Oct 6, 2026 · Role: Software Engineer Intern, 2027 summer
Two 45-minute coding rounds.
- Round 1:
record(timestamp)andgetHits(timestamp)for the past five minutes, with a queue or time buckets that clear old data. Follow-ups: any window length, and data too big for one machine: Five-Minute Hit Counter. - Round 2: given pairs where a trusts b, find who everyone else trusts and who trusts nobody, or -1. Follow-ups: trust passing along chains, and more than one judge: Find the Town Judge.
04: Two people, one destination, fewest edges
Report date: Oct 3, 2026 · Role: Summer Intern, 2027
One graph task; finished code wasn't required, a clear explanation was.
- In an unweighted, undirected graph, Alice starts at A, Bob at B, both go to D; use as few distinct edges as possible across both paths. The paths meet at some X and share X to D, so search from A, B and D and take the smallest sum of the three distances over X (X = D means no sharing). Ask first: is the graph connected, can A, B and D coincide, what if D is unreachable? Follow-ups were not listed. Practise with Minimum Union of Two Routes.
New grad: Software Engineer
Questions we found
| Question | Type | Last seen |
|---|---|---|
| Reach a Destination Through Scheduled Flights | Coding | Oct 2026 |
| Design a Safe Review Summarization Platform | System design | Oct 2026 |
Timeline
05: A review-summary pipeline and timed-trip reachability
Report date: Oct 8, 2026 · Role: Software Engineer (title says L4) · Level: New grad
The report names two tasks, with no round order or result.
- System design: a safe pipeline that ingests customer reviews, writes language-model summaries and feeds an internal dashboard: Design a Safe Review Summarization Platform.
- Coding: reach a place by chaining scheduled train legs with set times; ours uses flights: Reach a Destination Through Scheduled Flights.
Experienced full-time: Software Engineer
Questions we found
Dates are report dates, except where a report gave its interview dates.
| Question | Reports | Last seen |
|---|---|---|
| Count Good Numbers | 1 | Oct 2026 |
| Finish Time of Fixed-Length CPU Tasks | 1 | Apr–Jun 2026 |
| Lowest-Bid Auction System | 1 | Apr–Jun 2026 |
| Minimum Product Path in a Directed Graph | 1 | Apr–Jun 2026 |
| BQ: another team already built your project | 1 | Apr–Jun 2026 |
Timelines
06: Counting "perfect wake" numbers
Interview date: October 2026 · Report date: Oct 29, 2026 · Role: Software Engineer (title says SWE III) · Level: Experienced, changing jobs · Result: Not known
The candidate came in by referral and interviewed by video. The interviewer was encouraging throughout.
- Round 1, coding: a "perfect wake number" has no 0, no repeated digit, and passes a rule on each digit and its neighbors that the report words unclearly (its one example says 123 fails). The candidate brute-forced
isPerfectWakeNumber(n). - Follow-up: how many lie in 1 to 9999? The interviewer dropped the neighbor rule to make it easier. They thought of subset backtracking but didn't finish, and said nerves got in the way. Count Good Numbers practises the count with the rule "no digit lower than both neighbors", so 123 passes there.
The behavioral round was moved to a later date; they didn't expect to pass.
07: Product-cost paths, CPU tasks, an auction class, a downlevel
Interview date: April–June 2026 · Report date: Oct 27, 2026 · Role: Software Engineer, interviewed for L4 · Level: Experienced, about 2 years · Result: Passed at L3
After many applications a recruiter reached out. After a personality test, a contract recruiter handed them to a U.S.-based recruiter.
- Round 1: a Dijkstra variant where a path costs the product of its edge weights, not the sum; find the smallest: Minimum Product Path in a Directed Graph. Behavioral questions too; the hard one was finding, halfway through, that another team already built your project (practice).
- Round 2 (Bay Area onsite): a meeting-rooms twist. Given start times such as {1, 4, 6}, one run time shared by every task and a CPU limit, when does the last task end? Finish Time of Fixed-Length CPU Tasks. Then a lowest-bid auction on a heap with bid and offer calls, which they called very simple. Follow-ups: bids in a known range (counting sort), and removing an auction (a set check). See Lowest-Bid Auction System.
Weeks after the hiring committee, the result came back at L3 with no appeal, before any team match. The only way on was L3 team matching.
Cohort not reported
Questions we found
| Question | Type | Last seen |
|---|---|---|
| Fuel Route with Fewest Station Visits | Coding | Oct 2026 |
| Schedule Multiple Software Timers | Coding | Oct 2026 |
| Design a Pull-Based Server Metrics Monitor | System design | Oct 2026 |
| Someone takes credit for your work | BQ | Oct 2026 |
| No idea how to proceed on a project | BQ | Oct 2026 |
| Your career disappointment | BQ | Oct 2026 |
Timelines
08: A fuel route with forced refuels
Report date: Oct 27, 2026 · Role: Forward Deployed Engineer · Result: Not given
The candidate felt this coding round went badly.
- N cities and M roads, each road joining two different cities and burning 1 kg of fuel; the car starts at the source with P kg. Stopping in a station city forces a refill to exactly P, and the source is never a station. Roads can repeat, and each use counts. In priority order: fuel never goes below zero, then the fewest stations, then the fewest roads. Return both counts, or say the trip is impossible: Fuel Route with Fewest Station Visits.
09: Googleyness questions and timers on one hardware timer
Report date: Oct 23, 2026 · Level: L3 (the post names no role)
- Googleyness and Leadership: someone takes credit for your work (practice); no idea how to proceed on a project (practice); your career disappointment (practice).
- Coding, one question: a small embedded system has one hardware timer that fires an interrupt handler after a relative timeout. Only one can be pending, and setting it again replaces the deadline without firing. A clock gives milliseconds since boot. Build any number of timers on top, each with its own callback. The caller allocates each timer struct and never touches it, so you may add fields; callbacks run inside the interrupt. Write both the set-timer call and the handler: Schedule Multiple Software Timers.
10: A pull-based server metrics monitor
Report date: Oct 16, 2026 · Round: Second round, system design
- Every 10 minutes, pull CPU, memory, disk and app-health metrics from 1,000 servers, store them, and drive dashboards and alerts, growing to 100,000+ servers. The interviewer wanted real code for the worker that collects in parallel.
- The report's suggested design: one run per time slot even when several scheduler copies fire (a unique database key or a leader lease), and a set rule for when the last run is still going (skip, replace or run both). The worker caps parallel calls, times out each server on its own, and keeps going when one is down. Metrics go to a time-series store, and a failed pull shows as unknown, never as 0.
- Follow-ups: how the system behaves when the network drops, and how to debug it at 2 a.m. (logs, metrics, a traceable id per run). Practise with Design a Pull-Based Server Metrics Monitor.
Common questions
How many rounds is the Google SWE intern onsite?
Two of 4 intern reports described two coding rounds (02, 03); 02 and 03 gave 45 minutes each, and 02 had both on one day.
Does the Google intern onsite ask behavioral questions?
One intern (02) said there were none; the other intern reports did not mention any.
Can you run code in Google's interview editor?
One intern (02) said no: it was plain text, so they traced their code by hand.
What does a Google downlevel after the onsite look like?
One experienced candidate (07) interviewed for L4, got L3 from the hiring committee with no appeal, and could go on only through L3 team matching.
Keep practicing
- Google coding bank: every coding problem above and more (sign-in needed).
- Google system design bank: the designs from 05 and 10.
- Google behavioral bank: the questions from 07 and 09.
- Google interview guide and hiring insights.
- Experience feed: more candidate reports.
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.