Research Scientist, Gemini Multilinguality, Google DeepMind

Google•Mountain View, CA

17h•$207,000 - $300,000

About The Position

Google DeepMind is a pioneering AI lab with interdisciplinary teams advancing AI to address complex global challenges and accelerate product innovation. The lab prioritizes safety and ethics and uses its technologies for public benefit and scientific discovery. This role drives research to improve Gemini's multilingual reasoning, tool use, factuality, and multimodal performance, with an emphasis on people across languages, dialects, and locales. It collaborates with research and product teams to bring research advances into flagship Gemini models.

Requirements

Master's degree in Computer Science, Machine Learning, Artificial Intelligence, Mathematics, or a related technical field, or equivalent practical experience.
Research experience in one or more of the following areas: LLM post-training and alignment, reinforcement learning, multilingual benchmarking, reasoning and agentic systems, or foundation model training.
Experience training, post-training, or evaluating foundation models.

Nice To Haves

Ability to translate open-ended research ideas into landed improvements in frontier models across multi-team collaborations.
Excellent communication and collaboration skills.
Genuine passion for human languages and building AI that works for people everywhere.

Responsibilities

Drive research in post-training, reinforcement learning, and agentic modeling to improve Gemini's reasoning, tool use, factuality, and multimodal performance across global languages, dialects, and locales.
Develop scalable algorithms for cross-lingual capability transfer, synthetic data generation, reward modeling, and self-improvement to improve quality for major and low-resource languages.
Improve local and cultural grounding, factuality, naturalness, and language consistency in long-horizon, multi-turn, multimodal interactions involving text, speech, images, and video.
Design trustworthy evaluation methodologies and human-grounded reward signals to measure and steer complex agentic and culturally nuanced model behaviors.
Partner with Gemini post-training, agentic, and pretraining/mid-training teams, as well as Google product partners, to bring research breakthroughs into flagship Gemini models.