Google · Phone screen → Onsite
In processGoogle SWE III ML Experienced Loops: 10 Experiences (Part 3 of 4)
Google SWE III (L4) ML-track loops for experienced hires, March 2024 to May 2025: two or three coding rounds, an ML design round and G&L; four offers.
At a glance
- Process: phone screen → two or three coding rounds → an ML domain round (design or case study) → Googleyness & Leadership (G&L) → hiring committee and team matching (entries 19–28).
- Asked recently: an ML design or applied case in 6 of 10, such as email retrieval and ranking with a user profile (entry 19); string problems in 3 of 10; G&L in 6 of 10.
- Difficulty: mostly medium coding with hard follow-ups (entries 21, 28); one candidate chose an ML round that turned out to be theoretical statistics (entry 23).
- Practise first: group rotation-equivalent strings, file system total size and stable top K frequent words.
Coverage: 10 experienced-hire reports for Software Engineer III / L4 ML-track roles, March 2024 to May 2025. Hyderabad, a Bengaluru role, other parts of India and the US. Four offers, two rejections.
In this series
- Part 1: L4 ML-track loops from fall 2026 back to January 2026 (entries 01–10), with the topic summary for all 28 ML-track reports.
- Part 2: ML-track loops from November 2025 back to June 2025 (entries 11–18), one with a Gemini team-match round.
- Part 3 (current): 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.
Candidates who did not state their work experience have their own series: Google SWE III ML, hiring level not reported.
What came up
Six of these ten candidates described an ML design round, and six (not the same six) a G&L round. Strings were the most common coding theme. Only one question, going beyond your role, came from more than one candidate.
| Topic | Reports | Last seen |
|---|---|---|
| Googleyness & Leadership round | 6 of 10 | May 2025 |
| ML system design or applied ML case | 6 of 10 | May 2025 |
| Strings and words | 3 of 10 | May 2025 |
| Retrieval, ranking and recommendation | 2 of 10 | May 2025 |
| Hash maps, heaps and counting | 2 of 10 | May 2025 |
| Trees | 2 of 10 | May 2025 |
| Streams and sliding windows | 2 of 10 | May 2025 |
| ML theory and fundamentals | 2 of 10 | Feb 2025 |
Graphs (19), arrays and sorting (20), dynamic programming (21), statistics (23), and binary search and GPUs (26) came up once each.
Questions we found
Algorithms / Data Structures
| Question | Reports | Last seen |
|---|---|---|
| Group strings that are rotations of each other | 1 | May 2025 (report date) |
| Deletion order for parent-child storage parts | 1 | May 2025 (report date) |
| Minimum teleporter repair days | 1 | May 2025 (report date) |
| Minimum hand raises, five-key span | 1 | Apr 2025 |
| Total size under a file system node | 1 | Apr 2025 |
| Top K frequent words | 1 | Apr 2025 |
| Dictionary prefix search | 1 | Apr 2025 |
| Dice outcomes where die A beats die B | 1 | Apr 2025 |
| Delete N-ary tree nodes, return the forest (close to LeetCode 1110) | 1 | Mar 2024 |
| Sliding-window trimmed average (LeetCode 1825 is an easier version) | 1 | Mar 2024 |
ML design and modeling
| Question | Reports | Last seen |
|---|---|---|
| Retrieve and rank emails using a user profile | 1 | May 2025 (report date) |
| NLP clustering design (related case) | 1 | Apr 2025 |
| Classify photos of about 500 tree species | 1 | Jul 2024 |
| Video recommendation case (practice) | 1 | May 2024 |
ML theory and statistics
| Question | Reports | Last seen |
|---|---|---|
| Theoretical statistics: distributions, estimators, lesser-known theorems | 1 | Feb 2025 |
| Timed ML concepts quiz (about 20 minutes) | 1 | May 2024 |
Behavioral (BQ)
| Question | Reports | Last seen |
|---|---|---|
| Going beyond your role | 2 | May 2025 |
| Harsh feedback that helped someone grow | 1 | May 2025 (report date) |
| What you want to learn next year | 1 | May 2025 (report date) |
| Your impact on the team; a peer conflict; self-introduction | 1 | May 2025 |
| Ambiguity, culture fit, deadlines, disagreeing with a manager, team conflict | 1 | Apr 2025 |
10 detailed candidate timelines
G&L is Googleyness & Leadership; HC is the hiring committee.
19: Hyderabad MLE L4 loop that cleared every round
Report date: May 23, 2025 · Level: L4 · Location: Hyderabad · Result: Cleared every round; team matching (May 26, 2025)
A phone screen, then four onsite rounds over about two weeks, starting 20–25 days later.
Phone screen
- Group strings that are rotations of each other; the example had four strings, one ending up alone.
Coding 1
- Deletion order for parent-child storage parts: only a part with no children can be deleted. Follow-up: delete parts in batches.
Coding 2
- Minimum repair days between countries joined by teleporters: a broken teleporter costs one day, a working one nothing.
G&L
- Going beyond your role; giving someone harsh feedback that helped them grow; what you want to learn next year.
ML domain
- Retrieve and rank emails for a query using a user profile, drawn from the candidate's past work: design first, then parts of it in code, then a personalization follow-up.
20: five coding problems and an offer
Interview date: onsite coding April 2025, G&L around mid-May 2025 · Level: L4 · Location: a Bengaluru role (interviewers in a US time zone) · Result: Offer
Phone screen
- Count dice outcomes where die A beats die B; faces are arbitrary and can repeat. Follow-up: both arrays sorted.
Onsite 1
- Minimum hand raises over keys with a five-key span. Follow-up: the thumb position for each segment.
Onsite 2
- Total size under any node of an id-to-entity file system. Follow-ups: adding files, and a timeout at boot.
Onsite 3
- Top K frequent words in a parsed document (follow-ups: streaming input, K = 0) and dictionary prefix search (follow-up: empty prefix).
Onsite 4, G&L
- Self-introduction, team impact, a peer conflict, extra effort for users beyond the role, plus ML depth on the candidate's own work. There was no separate ML round.
21: two coding rounds and an NLP clustering design, rejected
Interview date: around April 2025 (report date July 19, 2025) · Level: L4 · Result: Rejected about a week after the loop
- Phone screen: not described.
- Coding: a heap-based problem (not shared).
- Coding: a hard DP problem with a trie follow-up.
- G&L (25–30 minutes): ambiguity, culture fit, deadlines, disagreeing with a manager, team conflict, working with others.
- ML domain: a clustering design in NLP (prompt not shared); related practice: group review topics and evaluate their summaries.
22: NDA problems and an offer
Interview date: G&L April 7, 2025 · Level: L4 · Location: India · Result: Offer (May 21, 2025)
- Phone screen, two coding rounds and an ML system design with follow-ups; the problems were under NDA.
- Team match approved April 23; the packet was in HC review on May 5.
23: an ML round that turned out to be statistics
Interview date: onsite finished February 25, 2025 · Level: L4 · Result: Pending
- Three onsite rounds: not described.
- An ML round, chosen instead of a coding round: theoretical statistics on distributions, estimators and lesser-known theorems.
24: four team-match calls and no match
Interview date: by the end of February 2025 · Level: L4 · Location: US · Result: No team match by April 17, 2025
- Two coding rounds, ML and G&L; no task details.
- Four team-match calls: one SWE team and ML teams including YouTube Recommendation, AR and Search Ads.
25: a backend engineer's loop with an add-on ML round
Report date: January 14, 2025 (add-on ML round at least a month earlier) · Level: L4 · Result: No response for over a month
- Onsite composition unknown; an add-on ML round, not described.
26: a tree-species case, rejected
Interview date: first call April 7, 2024; coding about two months later; ML round about ten days after (around June–July 2024) · Level: L4 · Result: Rejected by July 23, 2024; six-month wait to reapply
- Coding: a medium-hard binary-search problem; a string problem with no follow-up; a design-flavored coding question with follow-ups.
- ML domain: classify photos of about 500 tree species as a product: approach, loss and metrics, then user experience, model readiness, and why GPUs train faster than CPUs beyond parallelism. No code. No G&L round.
27: an ML concept quiz and a video-recommendation case
Interview date: between March 16 and May 8, 2024 · Level: L4 · Result: Not reported
- Planned: two coding rounds, leadership and ML domain; only the ML round was described.
- ML domain: a ~20-minute concepts quiz from regularization up to optimizing deep networks, then a video recommendation case study; practice: recommend videos for a home feed.
28: Hyderabad L4 loop with an offer
Interview date: March 2024 · Level: L4 · Location: Hyderabad · Result: Offer, about a week and a half after G&L
- Recruiter call (15 minutes): background.
- Screening: a medium tree problem like LeetCode 1110 Delete Nodes and Return Forest; closest FastPrep match: delete N-ary tree nodes and return the forest.
- Onsite 1: an easier take on LeetCode 1825 Finding MK Average; FastPrep has the full version as sliding-window trimmed average.
- Onsite 2: a withheld medium tree problem with two hard follow-ups.
- Onsite 3: ML system design (not given). Onsite 4: general G&L questions.
Patterns across the ML track
These cover all 28 ML-track reports in Parts 1–3.
- Retrieval and ranking in ML design. Three candidates had a search or recommendation ML round: query-based document search (09), email retrieval with a user profile (19) and video recommendation (27).
- Levels moved both ways. Three loops began at L5 and ended at L4 (03, 09, 15); three ended in L3 terms (06, 07, 08).
- Extra rounds. Four candidates sat a round beyond the standard loop: GPU and technical add-ons (07), a coding retake (09), a retake after a voided round (17) and an add-on ML round (25).
- Follow-ups that changed the input or goal. Six candidates report one: rotatable envelopes (01), the fewest CPUs (11), a live window of recent events (12), a random next word weighted by frequency (18), deleting in batches (19), and sorted arrays and streaming input (20).
- Team matching took a day to a year. Entry 10 matched the day after two calls and entry 28 reached an offer within about a week and a half of G&L; entry 13 was still matching about a year later and entry 24 had four calls without a match.
Common questions
What is the Google SWE III ML domain round?
An ML design or applied case in most reports: email retrieval and ranking (19), NLP clustering (21), tree-species classification (26) and video recommendation with a concepts quiz (27). One candidate got theoretical statistics instead (23).
How many coding rounds are in the Google L4 ML loop?
Usually two or three after the phone screen (entries 19, 20, 21, 26, 28); entry 20 had three coding onsites and no separate ML round.
What G&L questions does Google ask L4 ML candidates?
Going beyond your role came up twice (19, 20); others were harsh feedback, peer and manager conflict, ambiguity and deadlines (19, 20, 21).
How long is Google team matching after the L4 loop?
It varied: an offer within about a week and a half of G&L (28), versus four calls and no match after two months (24).
Keep practicing
- Coding from these loops: rotation-equivalent strings, deletion order and teleporter repairs (19); dice pairs, piano hand raises, file system size, top K words and prefix search (20); N-ary forest and trimmed average (28).
- ML cases: personalized email search (19), review topic clustering (21), tree-species identification (26), home-feed video recommendation (27).
- 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.