Senior ML Engineer - Embodied AI Scaling Foundations

Start free →

Or apply yourself on the official page →

About the role

Job Description At General Motors, we're turning today's impossible into tomorrow's standard. Our vehicles already move millions of people every day, and we're building the autonomy that will drive them - L2 through L4, on real roads, at real scale. Making self-driving safe at that scale is one of the hardest AI problems there is. The Data Scaling team owns the data flywheel for AV foundation model pre-training and SFT. We determine what data the AV needs in order to learn driving behaviors at scale, and we define what data quality means across the loop. The team delivers ML models that move the product up the data scaling curves, turning better data composition into measurably better driving behavior. We work with the very large datasets GM already has and we define the next generation of highest-value datasets GM collects. With each major release we aim to 10x the effective data behind our models: more scale, more diversity, and more value extracted from every example. Why Join Us?     • Train on driving data almost nobody else has -  real-world miles from GM's fleets, plus synthetic sim data - scaling into billions of examples. Then decide which ones are worth it: ten thousand near-identical highway miles teach the model less than one unprotected left turn in the rain. Mixture design, curation, mining, and evaluation are how you find out which is which, working alongside other MLEs and research scientists. • Work on questions with no textbook answers yet. Scaling laws for language are well mapped by now; for embodied driving data - heavy-tailed, safety-constrained, closed-loop - they aren't. You'd be helping write them, and we support publishing what you find. • See your results in the world rather than on a leaderboard. The models this team ships change how the vehicle behaves on real roads, and that behavior comes back as the evidence for your next iteration. As a Senior AI/ML Engineer in the Embodied AI Data Foundations organization, you will be an individual contributor developing data-centric AI solutions that directly improve autonomous driving performance. You will design and run the data curation and model training recipes that produce models capable of safe, reliable behavior across diverse real-world scenarios, drawing on both real and synthetic data.   What You'll Do   • Design and run experiments that connect data composition to model behavior: dataset mixtures, sampling strategies, curricula, and scaling-law studies that tell us where to invest next. • Apply methods such as self-supervised pre-training, imitation learning, reinforcement learning, and foundation-model fine-tuning to driving behavior, trajectory generation, and perception tasks. • Develop data curation and mining methods - auto-labeling, deduplication, difficulty and uncertainty estimation, long-tail and out-of-distribution scenario discovery - to raise the value of every training example. • Define offline metrics and evaluations that actually predict on-road behavior, and use them to make model and data decisions from evidence rather than intuition. • Trace model failures back to their root cause in the data, then close the loop by specifying the data needed to fix them. • Train models at scale across large multi-GPU/multi-node datasets, partnering with platform teams on the pipelines and tooling this requires. • Collaborate with cross-functional teams to bring models into onboard driving systems, and document learnings and best practices along the way. • Follow relevant literature and bring promising advances into our recipes and evaluations. Your Skills & Abilities   Required: • Master's or PhD in Computer Science, Robotics, Machine Learning. • Strong ML fundamentals: you can design a clean experiment, pick the right baseline, re

View the official posting →