Research Scientist, Gemini Omni, DeepMind

Overview

As a Research Scientist at DeepMind, you will advance state-of-the-art generative AI technologies and scalable training methodologies. You’ll design and train multi-modal models, optimize architectures for distributed accelerator clusters, and translate breakthroughs into production-ready systems. You will conduct rigorous research, publish findings, and contribute to safety- and ethics-focused development. You will collaborate with interdisciplinary teams to drive high-impact, real-world AI solutions.

Responsibilities

• Optimize model architectures and training pipelines for distributed training on accelerators (TPUs/GPUs)
• Design and train state-of-the-art generative models across image, video, and audio domains
• Develop and deploy reinforcement learning methods (RLHF, DPO, policy gradient) to align model behavior
• Implement efficient inference strategies (quantization, pruning, distillation, speculative decoding, kernel optimization) to reduce latency and memory footprint
• Bridge research and production by turning breakthroughs into scalable, robust algorithms and benchmarks
• Set up large-scale experiments, evaluate methods, and publish results; contribute to internal and external research communities

Key requirements

• PhD degree in computer science, mathematics, applied statistics, machine learning or equivalent practical experience
• Experience with training generative models (LLM, image, video)
• Experience with ML frameworks (TensorFlow or JAX or PyTorch)
• Experience conducting research
• Publication record in AI conferences (NeurIPS, CVPR, ICCV, ICLR)

• independent researcher with strong judgment
• ability to manage deadlines and deliverables
• cross-functional collaboration
• distributed training across accelerators (TPUs/GPUs)
• diffusion models, flow matching, autoregressive architectures
• reinforcement learning methods (RLHF, DPO, policy gradient)

…