Google PhD New Grad ML SWE Rejection: Blocked Paths, Prefix Deduplication, ML Basics, and BQ

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

The candidate completed four rounds and was rejected by email about a week and a half later without hiring-committee submission.

ML basics

Topics included logistic regression, Naive Bayes, transformers, classification metrics, bagging, and boosting. The candidate selected general ML rather than a narrower domain.

Behavioral interview

Questions covered the dissertation, why the candidate was proud of it, and times they agreed or disagreed with an advisor.

Coding 1: shortest path around blocked nodes

Given a graph, start, end, and a set of blocked nodes, return the path length from start to end without traversing blocked nodes. The candidate solved the base task quickly with BFS.

Follow-ups covered edge cases, space optimization, and then allowing blocked nodes at an extra time cost. That cost was never defined, so only the unweighted base task is linked.

Practice: Shortest Path Node Count Avoiding Broken Nodes

Coding 2: prefix deduplication across lists

Given listA, listB, and k, delete elements from listB so its first k values do not overlap the first k values of listA.

The follow-up generalized to a list of lists plus distance d: each list's first k values must avoid duplicates with the previous d lists. The candidate completed only a draft after an interviewer hint.

Practice: Deduplicate Card-Pile Prefixes