Principal Research Engineer, AI Systems, DeepMind

Google•Mountain View, CA

1d•$307,000 - $427,000

About The Position

Google DeepMind is developing the Agera Kernel, a scalable engine for pre-computation of expertise that goes beyond standard long-context retrieval. It uses map-reduce-style processing, agentic exploration, and persistent memory states to help models ingest, structure, and understand complex domains before user interactions. Automated domain pre-computation and expert human calibration in Agera Studio are intended to produce specialized Expert Models that perform strongly on intricate, multi-turn tasks. The role partners with product and research teams across Google DeepMind and Google to create products powered by Agera. It spans the AI systems stack and combines exploratory prototyping with the development of clean, scalable software. The engineer leads complex engagements from inception through delivery, sets technical direction, bridges team roadmaps and constraints, and drives alignment through working prototypes.

Requirements

Bachelor's degree in a related technical field, or equivalent practical experience.
15 years of experience in engineering or a related function.

Nice To Haves

Experience with the modern LLM stack, such as agent architectures and harnesses, multi-agent swarms, post-training or fine-tuning or reinforcement learning, or large-scale LLM evaluation.
Experience in software engineering, systems infrastructure, or AI/ML modeling using Python and C++.
Experience with technical leadership and driving multi-team, cross-functional projects from inception to delivery.
Experience in information retrieval, search and ranking systems, clustering, or large-scale data processing infrastructure.

Responsibilities

Partner with product and research teams across Google DeepMind and Google to invent 0-to-1 features and products powered by Agera, identify where Agera creates leverage, and align teams around shared goals.
Develop across the full AI systems stack, including agent harnesses and isolated subagent swarms, inspectable semi-structured state, vectorized tool execution, custom retrieval indices and universal reranking, large-scale evaluations, human-in-the-loop Studio UIs, and fine-tuning or expertise distillation into weights and skills.
Balance rapid, exploratory prototyping with writing clean, scalable code.
Lead complex engagements end-to-end, bridge team roadmaps and constraints, set technical direction, and drive convergence through working prototypes.