Google PhD New Grad Applied ML Onsite: Calendar Availability and Online Four-Candidate Ranking

Role: Software Engineer, PhD New Grad — Applied ML
Result: Rejected after the onsite
Report time: April 2025
Application: Referral

The candidate completed four onsite rounds in early April after applying with a referral in February.

Behavioral round

Four or five questions focused on leadership and collaboration stories. The round ended after about 25 minutes.

Coding 1: parse user quantities from a file

Read every line of a file, extract useful fields such as a user and quantity, and aggregate them with a dictionary. Two string-processing follow-ups were asked, but the source does not preserve the file grammar or either follow-up, so no exact practice link is asserted.

Coding 2: group calendar availability

Given calendars for several people:

1. Return days when everyone is available.
2. Return days when at least P people are available.
3. Return continuous available periods of length X.

Practice: Calendar Availability for at Least P People

ML round: online four-candidate ranker

For each online event, choose one of four candidates, show it to the user, observe feedback, and improve future choices from that feedback.

The candidate struggled because they had prepared fixed ML templates and did not adapt quickly to the interviewer's framing. A commenter identified the family as a multi-armed-bandit problem.

Practice: Design an Online Four-Candidate Ranker

The candidate reported the rejection on April 29. Their main takeaway was to understand fundamentals flexibly rather than relying only on memorized templates.