Evals Engineer, Polaris (Fixed-Term Contract)
Google DeepMind · Mountain View, CA, USA
CHECKED 23M AGO
Frontier labonsite$350k–$525k
At a glance
Highlights
- Design in-depth eval tasks testing Gemini on complex software engineering scenarios.
- Build graders and rubrics that score model attempts on realistic long-horizon work.
- Evaluate Gemini weaknesses to develop data and pipelines for model improvement.
What they offer
- Sign-on bonus
- 15% target annual bonus
Ideal candidate
Analytical problem-solver who loves hands-on building with AI coding agents and thrives on complex puzzles. Drawn to crafting realistic, high-craft evaluations that push frontier models forward.
Requirements
- Experience using AI coding agents
- Strong analytical and problem-solving skills
- Experience participating in competitive programming, mathematics competitions (e.g., IMO, USAMO, Putnam, or IOI)
Lab profile
About Google DeepMind
Google's AI lab building frontier models like Gemini to advance science and general intelligence.
What they build
- Gemini
- Gemma
- AlphaFold
Known for
- Frontier foundation models and GenAI
- Large-scale training with JAX, PyTorch, TensorFlow
- Model evaluation and deployment expertise
- Responsible frontier AI research
Key facts
- Founded in 2010
- Headquartered in London, UK
- Part of Google (Alphabet)