Senior Product Manager, Gemini Post Training, Deepmind

Google Google · Big Tech · Mountain View, CA +1

Senior Product Manager for Gemini Post Training at Google DeepMind. This role involves guiding the post-training and evaluation process for Gemini models, translating user needs into model development priorities, and collaborating with researchers to analyze model outputs and ensure quality. The focus is on defining evaluation goals, curating evaluation suites, and investigating issues to improve model performance and readiness for release.

What you'd actually do

  1. Define key capabilities and translate them into measurable evaluation goals for each model release cycle.
  2. Curate and evolve the post-training evaluation suite to accurately gauge model quality, readiness, and performance.
  3. Anticipate measurement needs 2–3 release cycles ahead, partnering with teams to develop evaluations for emerging capabilities.
  4. Collaborate closely with researchers and training leads to analyze checkpoints, interpret results, and guide daily training runs.
  5. Investigate and resolve high-impact issues, such as output regressions and behavioral artifacts, by rallying cross-functional teams to implement fixes.

Skills

Required

  • product management
  • technical role
  • taking technical products from conception to launch

Nice to have

  • interpret model evaluations
  • identify regressions
  • understand performance changes beyond tracking dashboards
  • engage in substantive training discussions regarding data quality and trade-offs with researchers
  • grow in fast-paced, ambiguous environments with rapidly shifting priorities and roadmaps
  • judge outputs to spot subtle capability, formatting, or tonal issues
  • act as a bridge across teams inside and outside of a team to align priorities and drive action

What the JD emphasized

  • reading evaluations
  • analyzing model outputs
  • making judgment calls on quality
  • product intuition and technical depth
  • forming a point of view on model quality
  • grounding it in real user needs
  • advocating for it in a fast-moving, research-driven environment
  • interpret model evaluations
  • identify regressions
  • understand performance changes
  • engage in substantive training discussions regarding data quality and trade-offs with researchers
  • judge outputs to spot subtle capability, formatting, or tonal issues

Other signals

  • product management
  • AI models
  • evaluation
  • training