About the job
Research Engineer / Scientist, Response Quality/Hermes
Mountain View, CA
At Google DeepMind, we celebrate the diversity of experiences and perspectives, leveraging them to create a transformative impact. We are dedicated to providing equal employment opportunities to all, regardless of sex, race, religion, ethnicity, disability, age, citizenship, marital status, sexual orientation, gender identity, or any other legally protected status. If you require accommodations due to a disability or other needs, please let us know.
Team Overview
Join a visionary team.
At Google DeepMind, our multidisciplinary team of engineers and scientists is pivotal in advancing Gemini’s post-training evolution. Our goal is to evolve Gemini from a static answer engine into a dynamic, collaborative AI partner capable of rich, multi-faceted interactions.
Our initiatives encompass a comprehensive range of model alignment and interaction design, including:
- Advanced Modeling Techniques: Innovating and scaling SFT, RL, and Reward Modeling techniques to enhance multi-turn reasoning and collaborative behaviors.
- Next-Generation Evaluation: Developing autonomous user-agent evaluation methods to emulate complex interactions and measure the quality of human-AI exchanges.
- Multimodal Response Generation: Enhancing user experience by integrating diverse formats such as images, charts, and interactive widgets beyond traditional text responses.
About Google DeepMind
Artificial Intelligence is poised to be one of humanity’s most transformative inventions. At Google DeepMind, our team of scientists, engineers, and machine learning experts collaborate to push the boundaries of AI technology. Our commitment to safety and ethics ensures that we prioritize public benefit and scientific discovery in all our endeavors.
Role Overview
As a Research Engineer/Scientist within the Gemini Post-Training team, you'll play a crucial role in bridging the gap between core model alignment and next-gen interactive AI. You will reimagine how Gemini engages, reasons, and collaborates with users through complex, multi-turn exchanges.

