Technical Program Manager, Gemini Post-Training, Subagents, DeepMind

Google's projects, like our users, span the globe and require managers to keep the big picture in focus while being able to dive into the unique engineering challenges we face daily. As a Technical Program Manager at Google, you lead complex, multi-disciplinary engineering projects using your engineering expertise. You plan requirements with internal customers and usher projects through the entire project lifecycle. This includes managing project schedules, identifying risks and clearly communicating them to project stakeholders. You're equally at home explaining your team's analyses and recommendations to executives as you are discussing the technical trade-offs in product development with engineers.

Using your extensive technical and leadership expertise, you manage projects of various size and scope, identifying future opportunities, improving processes and driving the technical directions of your programs.

This role sits within the Gemini Post-Training team in the DeepMind GenAI unit, which researches and enables new capabilities across our Gemini models. Our team builds large language models end-to-end through SFT, RL, inference-time self-improvement and many other techniques.

As a Technical Program Manager for Gemini Post-Training (Subagents), you will apply program management methodologies across multiple post-training data, evaluation, harness, and RL initiatives, balancing rapid progress and risk to deliver improvements across Gemini. You will partner with researchers, engineers, product managers, and cross-functional stakeholders to define project scope, goals, timelines, and resource requirements, turning research priorities into high-quality training data, evaluations, and deliverable model milestones. Ultimately, you will enable subagent capabilities across Gemini Post-Training to improve both quality and efficiency across complex, long-running coding, research, and knowledge-work tasks.

Artificial intelligence will be one of humanity’s most transformative inventions. At DeepMind, we are a pioneering AI lab with exceptional interdisciplinary teams focused on advancing AI development to solve complex global challenges and accelerate high-quality product innovation for billions of users. We use our technologies for widespread public benefit and scientific discovery, ensuring safety and ethics are always our highest priority.

Responsibilities

• Work with research and engineering teams to optimize subagent initiatives spanning Supervised Fine-Tuning (SFT) distillation, Reinforcement Learning (RL), and dynamic harnesses into clear goals, milestones, and release plans.
• Partner with modeling and evaluation teams to monitor evaluation leaderboards, tracking latency, token usage, parallelization speedups, and quality improvements.
• Drive convergence between research training harnesses and production agent runtimes, managing cross-functional dependencies across data, evaluations, infrastructure, and compute resources.
• Contribute to technical proposals by analyzing multi-agent trajectories, sync vs. async delegation patterns, and overall performance headroom.
• Manage end-to-end project planning using effective workflows, and translate complex modeling progress into clear, decision-ready updates for leadership and cross-functional audiences.

Minimum qualifications:

• Bachelor's degree in a technical field or equivalent practical experience.
• 8 years of experience in program management.

Preferred qualifications:

• 8 years of program management experience driving delivery with research scientists and software engineers on foundation models, agentic systems, or RL pipelines.
• Knowledge of software and ML development lifecycles, compute capacity planning, and experience managing complex, cross-functional dependencies.
• Strong stakeholder management skills, with the ability to navigate ambiguity, influence outcomes, and align cross-functional teams without relying on direct authority.
• Analytical depth in evaluating agentic trajectories, diagnosing failure modes, and translating insights into concrete data, harness, or RL interventions.