RepostedGoogle logo

Google · Phone screen → Onsite

In process

Google PhD AI/ML SWE Coding and ML Domain Rounds: 1 Experience

Software engineer · New GradDoctorateNov 2025
Shared by

Google PhD early-career AI/ML SWE, 2025–26: top-K IPs with big-data follow-ups, a Basic Calculator II onsite and a deep ML round, then team match.

At a glance

  • Process: recruiter call (October 13) → ML domain chosen (October 16) → R1 coding and behavioral (November 24, passed December 8) → California onsite (February 5) → team match (February 25) (01).
  • Asked recently: top 10 IPs from log files with large-file, out-of-memory and distributed follow-ups; a Basic Calculator II variant; ML fundamentals from bias-variance to backpropagation (01).
  • Difficulty: the ML round tested depth and communication, not memorized definitions; the onsite wanted the whole plan on a whiteboard before code.
  • Practise first: Top-K IP Addresses from File Records and Basic Calculator II With Empty Input.

Coverage: one Google PhD early-career software engineer (AI/ML, applied ML) candidate; onsite in California; October 2025 to February 2026, report February 2026.

What came up

Topic Reports Last seen
Hashing and top-K 1 of 1 Nov 2025
Expression parsing 1 of 1 Feb 2026
ML fundamentals 1 of 1 Feb 2026

Questions we found

Algorithms / Data Structures

Question Reports Last seen
Basic Calculator II With Empty Input 1 Feb 2026
Top-K IP Addresses from File Records 1 Nov 2025

ML concepts

Question Reports Last seen
Why L1 regularization gives sparse weights 1 Feb 2026
Backpropagation: loss, chain rule, updates 1 Feb 2026
Bias-variance trade-off 1 Feb 2026
Logistic regression, gradient descent, cross-entropy 1 Feb 2026

1 detailed candidate timeline

01: An applied-ML track through team match

Report date: February 2026 · Role: Software Engineer, PhD Early Career (AI/ML) · ML area: applied ML · Location: California onsite · Result: advanced to team match

Timeline

  1. Recruiter call on October 13; picked the ML domain on October 16; R1 coding and behavioral on November 24; passed on December 8; onsite on February 5; positive feedback and team match on February 25.

R1 coding

  1. Given log files, return the ten most frequent IP addresses: Top-K IP Addresses from File Records. The candidate used a frequency map and a top-K structure.
  2. Follow-ups: what changes for a very large file, what if the data does not fit in memory, and how it would work in a distributed setup. The interviewer stayed on these longer because of the candidate's distributed computing and storage background.

Onsite coding

  1. A Basic Calculator II variant with +, -, * and /, plus extra behavior for empty input: Basic Calculator II With Empty Input.
  2. The interviewer expected the full approach on a whiteboard before coding, edge cases raised without prompting, clear time and space complexity, and a complete, consistent solution over a rushed one.

ML domain interview

  1. Topics: bias-variance trade-off, logistic regression, gradient descent, cross-entropy, L1 and L2 regularization, and forward and backward propagation.
  2. For L1, explain why it makes weights sparse and draw the constraint geometry. For backpropagation, explain the loss, how gradients flow by the chain rule, and how parameters update.

Common questions

What is in the Google AI/ML SWE domain interview?

In report 01, ML fundamentals with deep follow-ups: why L1 gives sparsity (with the geometry drawn) and how backpropagation works step by step.

What follow-ups come with the Google top-K IP question?

Report 01 was asked about very large files, data that does not fit in memory, and a distributed version.

How long does the Google PhD AI/ML process take?

Report 01 went from a recruiter call on October 13 to team match on February 25.

Keep practicing

Sources: Anywhere useful interview info can be found.

0 total reactions. Your reactions: none.
0 comments

Comments

0

No comments yet — be the first to share what you know.