Conduct fundamental and applied research on multimodal LLMs and agentic workflows for generative media, including multi-turn reasoning, planning, instruction following, and tool use. Design, prototype, and run experiments on post-training and alignment methods (such as fine-tuning, reinforcement learning, and synthetic data curation) to improve model quality, faithfulness, and controllability. Define data pipelines, benchmarks, reward models, and evaluation metrics for multimodal generation and interactive editing under minimal guidance, refining hypotheses through empirical analysis.
Identify and investigate research opportunities in multilingual, cross-lingual, and cross-script multimodal modeling, collaborating with internal teams to shape research strategy and expand impact. Author research papers to share findings in the research community, and partner with Engineering teams to transfer promising research into Google products.
Job description provided by Google.
View the original posting
