Google

Summary

I completed a 30‑minute hiring manager interview for a Software Engineer III (ML) role on Google’s Gemini Applications team in Bengaluru, discussing project experience and system‑design questions.

How it went

Google Hiring Manager Round (Gemini Applications Team)

Recently, I gave the Hiring Manager round for SWE III (ML) at Google for the Gemini Applications team.
I have ~4.5 years of experience as a Data Scientist in a mid-size company.
The interview lasted around 30 minutes and focused on introduction followed by questions around building and deploying scalable systems.

Interview Questions & Discussion

Q1. Can you introduce yourself and walk me through one project?

My Answer:
Introduction and brief discussion about one project

Q2. How is evaluation done for such systems?

My Answer:
Answered relevant to the project

Q3. How would you design a multi-task system that handles varying domains?

My Answer:
Given constraints like limited resources and business pressure, and assuming access to a new model from the Gemini team:

Use golden datasets to evaluate performance across different tasks
Apply prompt enhancements instead of fine‑tuning due to time constraints
Highlighted that a single model may not effectively handle all domains

Q4. How would you develop or improve the model over time?

My Answer:
With the help of golden datasets and stakeholders

Q5. What would you do if no additional data is available?

My Answer:
Could not give a satisfactory answer
(Expected: agentic‑based evaluation on top of golden datasets to iterate faster)

My Questions to the Interviewer

Q1. How do you ensure systems are reliable across so many domains?

Q2. Do you use internal frameworks, or is there a dedicated team building solutions for new agentic workflows while others focus on implementation?

Other Interviews:

Other Tech Firms

Compensation

Offer / Compensation Discussions

Questions asked (3)

01
Design multi‑task system for varying domains
System Design
How would you design a multi‑task system that handles varying domains? Constraints include limited resources, business pressure, and access to a new model from the Gemini team. Consider using golden datasets for evaluation, prompt enhancements instead of fine‑tuning, and the possibility that a single model may not effectively handle all domains.
02
Develop or improve the model over time
System Design
How would you develop or improve the model over time?
03
Handling lack of additional data
System Design
What would you do if no additional data is available?

Google | Hiring Manager Round | Gemini Applications | SWE III ML | Bengaluru