Nvidia · posted · checked
Senior Data Scientist, Data Quality and AI Evaluation
PythonSQL$224k – $357kOn-site · Santa ClaraSenior
Who can apply
On-site · Santa Clara
Location: US, CA, Santa Clara
At NVIDIA, we are building the next generation AI-native autonomous driving architecture — combining classical safety stacks, foundation models, and scalable AI systems into a unified production platform. Autonomous driving is entering a new era. Future AV systems will not be purely classical robotics stacks, nor purely end-to-end neural networks. The future lies in the deep integration of classical safety architectures and large-scale AI-driven systems. If you want to work on problems that build the future of autonomous systems over the next decade, this is the place.
What You’ll Be Doing
Define quality metrics and evaluation methods that reveal where auto-labeling pipelines succeed, fail, or regress.
Build and maintain human-verified golden datasets covering representative driving conditions, critical corner cases, historical failures, and new requirements.
Analyze large-scale sensor, camera and annotation data to identify root causes, discover systematic errors, and prioritize improvements with perception and auto-labeling engineers.
Develop automated approaches for failure clustering, similar-case discovery, and targeted selection of data for human correction.
Build statistically sound audits and experiments, establish release acceptance criteria, and communicate findings through clear dashboards and recommendations.
Partner with labeling operations to improve annotation guidelines, reviewer consistency, and the balance between quality, efficiency, and cost.
What We Need To See
Great foundations in statistics, experimental design, sampling, hypothesis testing, and uncertainty estimation, with practical experience evaluating machine learning systems and data quality.
Experience working on AI/LLM evaluation with proficiency in Python and SQL, including analyzing large datasets and building reproducible analytical workflows. Understanding of precision, recall, class imbalance, selection bias, data leakage, and evaluation across significant scenario groups.
Ability to distinguish reference-label defects from model errors and translate ambiguous quality concerns into measurable questions and actionable recommendations.