Senior Research Engineer, Multimodal Foundation Models (DeepMind)

About Google DeepMind

Google DeepMind is committed to solving intelligence to advance science and benefit humanity. Our team develops state-of-the-art multimodal foundation models like Gemini, uniting research breakthroughs in vision, audio, text, and code into unified, highly capable reasoning architectures.

Role Overview

As a Senior Research Engineer, you will bridge foundational scientific discoveries and production engineering. You will architect scalable training pipelines on Google Cloud TPU v5e/v5p pods and develop novel neural architectures that unlock emergent reasoning across multimodal modalities.

Key Responsibilities

• Drive experimental iterations on next-generation Gemini vision-language-audio pre-training recipes and loss formulations.
• Implement efficient attention mechanisms, linear attention variants, and memory-efficient sequence models in JAX/Flax.
• Build robust evaluation harnesses to benchmark multi-turn agentic capabilities, spatial reasoning, and code synthesis.
• Collaborate with product teams across Google Cloud and Search to deploy optimized, low-latency distilled model variants.

Qualifications & Experience

• Master's or PhD degree in Computer Science, AI, or equivalent quantitative discipline.
• 5+ years of software engineering and machine learning experience with strong proficiency in Python and C++.
• Extensive experience with distributed ML frameworks (JAX, TensorFlow, PyTorch) across TPU or GPU accelerators.
• Demonstrated ability to debug numerical instabilities, gradient explosions, and hardware checkpointing in massive models.

Benefits & Total Rewards

• Base Salary: $210,000 – $325,000 per year + annual corporate bonus target.
• Google Equity (Restricted Stock Units - GSUs) with flexible vesting schedules.
• Industry-leading 401(k) 50% match up to IRS statutory limit.
• On-campus wellness facilities, medical clinics, childcare subsidies, and global sabbatical programs.