Google · Phone screen → Onsite
In processGoogle L4 ML Loops and Team Matching: 8 Experiences (Part 2 of 4)
Google SWE III (L4) ML-track loops for experienced hires, January to November 2025: coding, ML rounds and Googleyness, with long team matching.
At a glance
- Process: phone or screening round → two or three coding rounds → an ML round (design, theory or, once, coding) → Googleyness & Leadership → team matching and HC (entries 11–18).
- Asked recently: CPU task scheduling and unlock patterns on a dot grid (11); a most-frequent message in an event window (12); next-word prediction and fence painting (18).
- Difficulty: medium to medium-hard coding with follow-ups (14); one ML round had no design part, only theory (14), and one was run as coding (17).
- Practise first: Weighted next-word prediction, Shortest shared route and Most frequent events in a window.
Coverage: 8 experienced-hire reports for Software Engineer III / L4 ML-track roles (SWE III ML, SWE III AI/ML, MLE), one aiming for L5. India is the only location named. Interviews from January to November 2025; one team-match round in April 2026.
In this series
- Part 1: L4 ML-track loops from fall 2026 back to January 2026 (entries 01–10), plus the topic summary for all 28 ML-track reports.
- Part 2 (current): ML-track loops from November 2025 back to June 2025 (entries 11–18), one with a Gemini team-match round.
- Part 3: ML-track loops from May 2025 back to March 2024 (entries 19–28), then patterns across the ML track.
- Part 4: Forward Deployed Engineer and Cloud AI Engineer loops.
What came up
Five of these eight reports include a behavioral round, and three describe an ML design or applied-ML case. Four candidates named or described their coding problems (#11, #12, #14, #18).
| Topic | Reports | Last seen |
|---|---|---|
| Behavioral / Googleyness & Leadership round | 5 of 8 | Nov 2025 (report date) |
| ML system design or applied ML case | 3 of 8 | Nov 2025 (report date) |
| Hash maps, heaps and counting | 2 of 8 | Aug 2025 |
One report each covered backtracking (#11), streams (#12), ML theory and attention (#14), and graphs, strings and arrays (#18). A candidate can sit in several rows.
Questions we found
These tables cover the eight reports in this part. Each question comes from one candidate.
Algorithms / Data Structures
| Question | Reports | Last seen |
|---|---|---|
| Minimum time to finish M tasks on N CPUs | 1 | Nov 2025 (report date) |
| Count unlock patterns on an N × M dot grid | 1 | Nov 2025 (report date) |
| Most frequent message in a timestamped event window | 1 | Jul 2025 |
| Next-word prediction with a weighted random follow-up | 1 | Jun 2025 |
| Fewest strokes to paint a fence (Codeforces 448C) | 1 | Jun 2025 |
| Shortest shared route, an LC 2203 variant | 1 | May 2025 |
ML Design & Modeling
| Question | Reports | Last seen |
|---|---|---|
| Evaluate a multi-domain model with no new data (team-match round) | 1 | Apr 2026 |
| Assign a query integer to one of two arrays | 1 | Nov 2025 (report date) |
ML Theory & Fundamentals
| Question | Reports | Last seen |
|---|---|---|
| ML, NLP and attention theory next to a work-project deep dive | 1 | Jul 2025 (report date) |
Behavioral (BQ)
| Question | Reports | Last seen |
|---|---|---|
| Leading an overwhelmed team | 1 | Aug 2025 |
| Everything marked P0 | 1 | Aug 2025 |
| Your biggest disappointment | 1 | Aug 2025 |
| A peer taking credit on a four-person project | 1 | Aug 2025 |
| Helping a colleague fit into a diverse team | 1 | Aug 2025 |
| Owning up when you couldn't handle all your tasks | 1 | Aug 2025 |
8 detailed candidate timelines
G&L is Googleyness & Leadership and HC is the hiring committee.
11: L4 AI/ML loop with CPU scheduling, unlock patterns and a 1-D classification case
When: report 2025-11-23 · Level: L4 · Outcome: pending team matching, then HC
- Phone screen: minimum time to finish M tasks on N CPUs, one task per CPU at a time, no cooldown; follow-up: the fewest CPUs that still reach that minimum time. The practice version adds a fixed task order, which the report doesn't state.
- Coding: count unlock patterns of length L or more on an N × M dot grid: start on any dot, move to any of the 8 unvisited neighbours, no jumping over a visited dot; follow-up: a linear-factor speed-up.
- AI/ML: assign a query integer to one of two integer arrays.
- G&L (~30 min): behavioral questions, not listed.
12: SWE III ML loop in Bengaluru, followed by a Gemini team-match round
When: first round the first week of Jul 2025; G&L ~2025-08-10; team-match hiring-manager round ~Apr 2026 · Level: L4 · Where: Bengaluru · Outcome: not reported
- First round, coding (45 min): most frequent message among timestamped events, with its timestamps; each input line is
message: timestamp; follow-up: a real-time window of the last N events, oldest evicted, with the most frequent message after each arrival. - Coding (online, Google Docs): not written up.
- Domain: not written up.
- G&L: six themes: leading an overwhelmed team as a manager, everything marked P0, the biggest disappointment, a peer taking credit on a four-person project, helping a colleague fit into a diverse team, and owning up when unable to handle all of one's tasks.
- Team-match hiring-manager round, Gemini Applications (~30 min, after all four loop rounds): a self-intro and one project; how such systems get evaluated; then one multi-task system across many domains. Follow-ups: improve it over time, then do it with no extra data.
13: SWE III AI/ML loop in India with about a year of team matching
When: interviews Apr–Aug 2025, Googleyness last · Level: L4 · Where: India (Bangalore requested) · Outcome: cleared the loop; still in team matching on 2026-10-20
- Recruiter call: DSA questions with no coding.
- Mock DSA: a practice coding round.
- Coding ×3: problems not named.
- ML core (system design): prompt not reported.
- Googleyness: not described.
- Team-match interview (Dec 2025): an L4 AIML team round; whether it took place isn't clear.
An early team match fell through.
14: SWE III L4 MLE loop with an ML round that had no design part
When: report 2025-07-10 (order of the three onsite rounds not given) · Level: L4 · Where: India · Outcome: unknown
- Screening: not described.
- Onsite coding: a medium problem; follow-ups: optimize it, and write an iterative version.
- Onsite coding: a medium-hard problem.
- Onsite ML: a deep dive on a work project plus theory on ML, NLP and attention; no design part.
- Googleyness: pushed back a week; not reported.
15: MLE loop that targeted L5 and chose L4 team matching
When: report 2025-06-30 · Level: targeted L5 · Outcome: pending: chose L4 team matching over an L5 retry in a year
- Onsite coding, ML and behavioral rounds: no task details; the round count isn't given.
16: SWE 3 ML loop in India, in team matching with an extra round possible
When: phone screen late Mar 2025; ML and Googleyness the week of 2025-06-16 (inferred) · Level: L4 · Where: India · Outcome: in team matching on 2025-07-11; an extra round may be added
- Phone screen (late Mar 2025): not described.
- Coding 1, coding 2, an ML round (a design round) and Googleyness: no task details.
17: SWE III AI/ML loop where the ML slot was run as coding
When: screening Jan 2025; onsite Apr–Jun 2025, latest round ~2025-06-10 (inferred) · Level: L4 · Outcome: ML round and Googleyness still pending on 2025-06-12
- Screening (Jan 2025): not described.
- Onsite 1 (Apr 2025): voided after a shared-doc glitch moved it to a Google Doc.
- Onsite 1 retake (May 2025): not described.
- Onsite 2 (May 2025): not described.
- ML slot, run as coding (~2025-06-10): a problem-solving coding question.
The ML round itself was rescheduled to ~2025-06-18 (inferred).
18: MLE loop with three onsite rounds, rejected before Googleyness
When: recruiter ~early Apr 2025; phone ~May 2025; onsites ~Jun 2025, within one week (inferred) · Level: L4 · Where: India · Outcome: rejected; no Googleyness round took place
- Phone screen (~35 min): shortest shared route, an LC 2203 variant on an undirected, unweighted graph; follow-up: three friends meeting at a fourth.
- Onsite 1, coding: next-word prediction API over sentences: return the most frequent follower, with many edge cases; follow-up: pick the next word at random, weighted by frequency (close to LC 528).
- Onsite 2, coding (35-minute limit): fewest strokes to paint a fence (Codeforces 448C).
- Onsite 3, ML: a non-standard, research-style problem, prompt withheld.
Common questions
What coding questions did Google ask L4 ML candidates in 2025?
CPU task scheduling and grid unlock patterns (11), a most-frequent-message window (12), a shortest shared route (18), next-word prediction (18) and fence painting (18).
What happens in a Google team-match round?
In entry 12, a 30-minute hiring-manager round for Gemini Applications: a project, how such systems are evaluated, and an ML case on evaluating a multi-domain model with no new data.
How long can Google team matching take?
About a year in entry 13, which cleared the loop in 2025 and was still matching in October 2026.
Keep practicing
- Coding from this part: CPU task finish time and unlock patterns (11); events in a window (12); shared route, next-word prediction and fence painting (18).
- ML cases from this part: one-dimensional group membership (11) and multi-domain model evaluation (12).
- Google question banks: algorithms and system design (sign-in needed).
- Everything Google in one place: the Google interview page and Google hiring insights.
- More firsthand reports: the Experience feed.
In this series
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.