Senior AI/ML Engineer
$171k - $261k • Sunnyvale, California, United States of America • Full-time
Posted 10d ago
Job Location
Sunnyvale, California, United States of America
Tech Stack
Remote Work Policy
On-site
Employment Type
Full-time
Categories
Machine Learning Engineer
About the job
We are seeking a mathematically rigorous engineer to own model numerics, focusing on how we measure, bound, and reason about the numerical behavior of models and how this analysis informs deployment decisions. The core challenge involves determining if numerical differences between model versions are safe, connecting low-level concepts like floating-point drift to high-level outcomes like vehicle trajectory error. You will build the tooling to quantify these connections and define the thresholds for ship/no-ship decisions. This role requires more than just running existing validation; it demands the ability to investigate and explain the root causes of numerical divergence.
Responsibilities
- Validate optimized model implementations to ensure equivalence with their references.
- Connect tensor-level numerical differences to downstream driving behavior and identify the underlying mechanisms.
- Build tooling for sensitivity and robustness analysis using Jacobian/Hessian-based methods.
- Design diagnostics to detect and root-cause training instabilities.
- Develop efficient metric computation, gradient decomposition, and diagnostic logging for distributed training jobs.
Requirements
- Strong command of numerical analysis and matrix theory, including Jacobian/Hessian estimation, spectral properties, conditioning, and floating-point error analysis.
- A deep understanding of neural network training dynamics, including loss landscapes, gradient propagation, optimizer behavior, and common failure modes.
- An aptitude for identifying numerical edge cases and understanding their impact.
- High proficiency in PyTorch and Python, with a proven track record of building production-ready analytical tools.
- The ability to translate complex numerical results into actionable insights for engineers.
- Bachelor's, Master's, or PhD in Applied Mathematics, Control, Physics, Computer Science, Data Science, or a related quantitative field.