Google · Phone screen → Onsite
In processGoogle SWE III ML Loops in India: 10 Experiences (Part 2 of 3)
Google SWE III (L4) ML-track loops, level not reported, April 2024 to October 2025: coding, ML rounds that sometimes ran as code or math, and Googleyness.
At a glance
- Process: phone screen → two or three coding rounds → an ML domain round → Googleyness & Leadership → team matching (entries 12–21).
- Asked recently: working ML code: an n-gram language model (14) and a sentiment model's fit and predict (18); a Q&A bot or chatbot design (15, 20); grid BFS with an extra constraint (15).
- Difficulty: some ML rounds leaned on math (12, 18), and one billed as ML ran as coding (16).
- Practise first: N-gram language model, documentation Q&A chatbot and fair message-log truncation.
Coverage: Hiring level not reported. 10 Software Engineer III / L4 ML-track loops, April 2024 to early October 2025. Bengaluru and India in general are the locations named.
In this series
- Part 1: L4 ML-track loops from November 2026 back to mid-October 2025 (entries 01–11), plus the topic summary for all 21 ML-track reports.
- Part 2 (current): ML-track loops from early October 2025 back to April 2024 (entries 12–21), then patterns across the ML track.
- Part 3: Forward Deployed Engineer and Cloud AI Engineer loops.
Candidates who stated their work experience are in the other series: Google SWE III ML, experienced hires.
What came up
Seven of these ten reports include a behavioral round, and three describe an ML design round.
| Topic | Reports | Last seen |
|---|---|---|
| ML theory and fundamentals Q&A | 2 of 10 | Oct 2025 |
| Behavioral / Googleyness & Leadership round | 7 of 10 | Sep 2025 |
| ML system design or applied ML case | 3 of 10 | Sep 2025 |
| Coding with an ML twist | 2 of 10 | Sep 2025 |
| LLMs, attention and GenAI | 2 of 10 | Jul 2025 (report date) |
| Graphs, grids and shortest paths | 2 of 10 | Jul 2025 (report date) |
Single reports covered heaps, strings, dynamic programming and statistics (#18), and backtracking (#21). A candidate can sit in several rows.
Questions we found
These tables cover the ten reports in this part. One question came from two candidates; the rest come from one each.
Algorithms / Data Structures
| Question | Reports | Last seen |
|---|---|---|
| Grid BFS with an extra constraint | 1 | Jul 2025 |
| Fair truncation of a category-tagged message log | 1 | Jun 2025 |
| Maximize distance over N days with K energy | 1 | Apr 2024 (report date) |
ML Coding
| Question | Reports | Last seen |
|---|---|---|
| N-gram language model in Python | 1 | Sep 2025 |
| Fit and predict for a described sentiment model (practice version: fit and predict a naive Bayes sentiment model) | 1 | Dec 2024 (report date) |
ML Design & Modeling
| Question | Reports | Last seen |
|---|---|---|
| Q&A bot or chatbot design (filed under one FastPrep problem; the two reports word the prompt differently, and whether it was the same prompt isn't known) | 2 | Jul 2025 |
Statistics & Probability
| Question | Reports | Last seen |
|---|---|---|
| Derive a posterior probability from given conditionals, exact question withheld (combine token posteriors with naive Bayes) | 1 | Dec 2024 (report date) |
Behavioral (BQ)
| Question | Reports | Last seen |
|---|---|---|
| Challenges you faced | 1 | Dec 2024 (report date) |
| Collaboration | 1 | Dec 2024 (report date) |
| Your values | 1 | Dec 2024 (report date) |
10 detailed candidate timelines
HM is the hiring manager.
12: Google L4 onsite with two coding problems and a math-heavy ML round
When: onsite ~2025-10-01 (inferred: the week before a 2025-10-06 post) · Level: L4 · Outcome: pending; Googlyness not yet scheduled
- Onsite coding: two problems, not named.
- Onsite ML domain: mathematical ML questions.
13: SWE III ML loop in Bangalore with an extra coding round and a long team match
When: extra coding round done by 2025-09-30 · Level: L4 · Where: Bangalore · Outcome: in team matching for about six months as of 2026-04-07; final result unknown
- Phone screen: not described.
- Onsite coding (online): problem not given.
- Onsite ML (online): ML theory plus ML system design; questions not shared.
- Onsite Googleyness (online): standard questions.
- Additional coding: not described.
14: SWE ML loop with a message-log phone screen and an n-gram ML round
When: phone screen ~Jun 2025 (tied to this loop by author and timing); Googliness by 2025-09-13 · Level: L4 · Outcome: unconfirmed
- Phone screen: fair truncation of a category-tagged message log to at most K entries: one shared cap applies to every category, and proportional allocation is ruled out; edge cases K = 0, an empty log, K above the length.
- Coding 1 and coding 2: not described.
- ML round: an n-gram language model in Python, plus ML questions.
- Googliness: not described.
15: SWE III ML loop with a constrained grid BFS and a Q&A bot design
When: loop ~Jun–Jul 2025 (inferred from a 2025-07-28 post; the report lists one onsite algorithms round) · Level: L4 · Outcome: in team matching for possibly L4; final result unknown
- Phone screen: not described.
- Onsite algorithms: grid BFS with an extra constraint, which the candidate likens to LeetCode's obstacle-elimination problem; follow-up: more constraints.
- Onsite ML system design: a Q&A bot.
- Googleyness: not described.
16: SWE ML loop where the ML round ran as coding
When: ~late Jun 2025 (inferred from a 2025-07-01 post) · Level: L4 · Outcome: unknown
- Round billed as ML, run as coding: DSA questions.
- Coding ×2 (scheduled): not described.
17: SWE3 ML L4 loop with four undescribed rounds
When: report 2025-01-29 · Level: L4 · Outcome: unknown
- R1 coding, R2 coding, R3 Googlyness and R4 ML: no task details.
18: ML Software Engineer loop in India with a posterior derivation and ML coding
When: 2024, month unknown (report 2024-12-21; screening and R1–R5 were 45 minutes each) · Level: L4 · Where: India · Outcome: through to team-match calls; final result not reported
- Screening: a string task whose harder follow-up needed sub-linear time.
- R1, coding: a graph problem best solved with multi-source BFS.
- R2, coding: a hard 2D DP.
- R3, ML domain: derive a posterior of the shape P(+|No!) from P(+), P(+|!) and P(+|No), exact question withheld (combine token posteriors with naive Bayes); then implement fit and predict for a described mathematical sentiment model (practice version: fit and predict a naive Bayes sentiment model), with a follow-up on TF-IDF, BPE and deep-learning alternatives.
- R4, coding: a LeetCode-hard heap problem.
- R5, Googlyness: challenges, collaboration, values.
- Team-match rounds: résumé-focused calls with hiring managers.
19: SWE III ML loop in Bengaluru with two unlabelled onsite rounds
When: report 2024-10-14 · Level: L4 · Where: Bengaluru · Outcome: unknown
- Screen, two unlabelled onsite rounds (possibly coding), onsite ML and onsite Googlyness: no task details.
20: Google L4 ML domain round with a Cloud AI chatbot design
When: report 2024-05-06 · Level: L4 · Outcome: not reported; other rounds not reported
- ML domain: a system design question, a chatbot for Google Cloud AI; FastPrep files it under its Q&A-bot problem, and the prompt wording differs.
21: SWE III ML loop with an energy-planning problem and a backtracking problem
When: report 2024-04-30 (loop as listed: phone screen, two coding rounds, ML round, Googlyness) · Level: L4 · Outcome: not reported
- Phone screening: not described.
- Second round: maximize the distance covered over N days with K energy: a work day covers distance and costs 1 energy, a rest day restores 1 (example: 7 days, K = 3; the statement is from memory). A backtracking problem also came up (statement not recalled); which round it was in isn't clear.
Patterns across the ML track
These cover all 21 ML-track reports in Parts 1–2.
-
A slot billed as ML ran in another format. Multiple candidates report it: #11's AIML domain round ran as a system design, and #16's round billed as ML was run as coding.
-
Extra rounds came up in five reports. An Applied AI team's two extra coding rounds (#01, not yet held), two add-on coding rounds (#04), two HC-requested add-ons (#06), an HC-requested technical round (#09), and an additional coding round (#13).
-
Graph search ran through the coding rounds. Five reports describe it: graph path-finding (#04), transitively similar movies (#05), shortest path and the smallest maximum edge (#10), grid BFS with a constraint (#15), and a multi-source BFS problem (#18).
-
Some ML rounds leaned on math. Multiple candidates describe it: mathematical ML questions (#12) and a posterior derivation (#18).
-
Two chatbot designs under one FastPrep problem. #15 (a Q&A bot) and #20 (a chatbot for Google Cloud AI) are filed under the same FastPrep problem. The reports word the prompt differently, and whether it was the same prompt isn't known.
-
Team matching took from one call to about six months. #04 matched on its first and only team-match call. #07's application closed under four weeks into matching, with no feedback from teams. At the slow end, #08 had no hiring-manager calls by 2026-07-17, about three and a half months after the loop, and #13 was still matching after about six months.
Common questions
Does the Google ML round ask you to write ML code?
In two of these ten reports, yes: an n-gram language model in Python (14) and a sentiment model's fit and predict (18).
Is the Google ML round always ML design?
No: one ran as coding (16), and two leaned on math such as a posterior derivation (12, 18); three were ML design rounds.
What ML system design does Google ask at L4?
A Q&A bot or chatbot design came up twice (15, 20).
Keep practicing
- Coding from this part: grid BFS with obstacle elimination (15), fair message-log truncation (14) and boat energy max distance (21).
- ML coding and math: n-gram language model (14), naive Bayes sentiment fit and predict and combined naive Bayes posterior (18).
- Design: documentation Q&A chatbot (15, 20).
- 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.