Gemini Vision Research Scientist, DeepMind

Google•Los Angeles, CA

2d•$207,000 - $300,000

About The Position

Google DeepMind is seeking a Research Scientist to advance image and video understanding capabilities in large language and multimodal foundation models. The role involves designing and implementing model capabilities, conducting end-to-end experiments across pre-training and post-training, and building scalable infrastructure and data pipelines for multimodal training and evaluation. You will work with Research Scientists and software engineers to turn research prototypes into production-ready models, while balancing rapid prototyping with maintainability and generality. The work includes analyzing experiment metrics and multimodal benchmark failure modes to improve model quality. Google's research portfolio spans fundamental research, product innovation, product contributions, and infrastructure. Research Scientists set up large-scale tests, deploy promising ideas, manage deadlines and deliverables, and apply current theories to develop products, processes, and technologies. They also contribute to the research community by sharing and publishing findings, including work inspired by internal projects and collaborations with universities and technical institutes. Google DeepMind describes its mission as advancing AI to address complex global challenges and accelerate product innovation, with safety and ethics as priorities.

Requirements

At least 3 years of experience in software engineering or a related technical field.
At least 2 years of work experience in an applied research setting.
Experience in Artificial Intelligence or Machine Learning.
Experience developing autonomous agents, multi-agent workflows, or agentic frameworks, such as AutoGen, CrewAI, LangGraph, tool-use/function-calling, or multi-agent reinforcement learning.
Experience designing, pre-training, fine-tuning, or evaluating large language models and multimodal foundation models.
Contribution to research communities or efforts, including authorship of papers at conferences such as NeurIPS, ICML, ACL, or CVPR.

Nice To Haves

Master's or PhD degree in Computer Science, Artificial Intelligence, Machine Learning, or a related technical field.
Demonstrated ability to independently manage data generation, model training, and optimization pipelines.

Responsibilities

Design, implement, and scale state-of-the-art LLM core image and video understanding capabilities.
Drive end-to-end experimental workflows across pre-training and post-training.
Build robust, scalable infrastructure and data pipelines for high-quality multimodal training and evaluation benchmarks.
Partner with Research Scientists and software engineers to transition research prototypes into production-ready foundation models, balancing rapid prototyping, maintainability, and generality.
Analyze experiment metrics and failure modes across multimodal benchmarks to identify opportunities to improve model quality.