Google 2026 PhD New Grad AI/ML R1: LLM Fine-Tuning, VAE, and BQ

Role: Software Engineer, PhD New Grad — AI/ML
Result: Advanced toward an onsite R2
Report time: January 2026

R1 contained one behavioral interview and one ML-domain interview. The recruiter offered several domain tracks in advance based on the candidate's background; the candidate selected LLM / generative AI.

ML-domain interview

The session began with about ten minutes of resume discussion. The remaining questions covered ML fundamentals tied to the chosen track:

• Common ways to fine-tune an LLM
• Advantages and disadvantages of those fine-tuning methods
• Scenario-based choices among the methods
• Autoencoders
• Variational autoencoders (VAE)
• Vector-quantized variational autoencoders (VQ-VAE)

The source preserves concept areas and tradeoff discussion but not a bounded coding or ML-system-design prompt.

Behavioral interview

Questions included:

• The candidate's most interesting project
• Working with people who are dramatically different
• Learning from someone with a different background
• Helping another person whose work has higher priority than the candidate's own work
• Dealing with challenges

A commenter reported that R2 would be two back-to-back onsite rounds.