About the Role
If you're passionate about pushing the boundaries of AI and solving some of the most challenging problems in the field, this is your chance. Come join our Evals team, where with products like Everest and EvalHub, we empower GenAI model development across Google. In this role, you will lead the product goal for how Google evaluates AI Agents, ensuring we build the most capable and trustworthy agentic systems in the world.
Responsibilities
•Develop a first-principles understanding of what makes an evaluation task truly useful—optimizing for coverage, climbability (scaling with compute and data while preserving headroom), discriminability, statistical stability (tight confidence intervals), and zero contamination.
•Synthesize insights into a holistic product goal and roadmap for Eval Quality.
•Operate seamlessly across Product Area boundaries between Gemini, AI2, and other partner product areas.
•Grow in Gemini's dynamic, fast-moving, research-driven environment where priorities evolve rapidly and Product Managers earn trust by directly contributing to day-to-day execution.
Requirements
•Bachelor's degree or equivalent practical experience.
•8 years of experience in a product management, engineering, and technology strategy role.
•5 years of experience taking technical products from conception to launch (e.g., ideation to execution, end-to-end, 0 to 1, etc.).
Nice to Have
•Experience with LLM/GenAI evaluations, post-training/mid-training RL/SFT workflows, autoraters, or human-in-the-loop (HiTL) expert data collection.
•Deep intuition for evaluation science—understanding what separates a brittle benchmark from a high-signal, climbable, uncontaminated evaluation across specialized verticals (e.g., finance, law, medicine, math, science, and engineering).
•Exceptional written and verbal communication skills, with the gravitas to align Executive Team Leads, Research Scientists, and VP/SVP leadership across London and US time zones.
•Demonstrated track record of growing across Product Area boundaries, specifically balancing the changing, execution culture of a research organization.
Conditions
No conditions stated in the posting.
