Senior Deep Learning Engineer, Cosmo 3D Spatial
Core
Own the data engine and end-to-end training verification for 3D spatial reasoning and perception capabilities in NVIDIA's Cosmos platform, ensuring models learn geometry, motion, and physical context from large-scale data.
Role type
Senior Deep Learning Engineer (3D Spatial Data & Evaluation)
Builds
3D spatial reasoning models for robots, autonomous vehicles, and smart infrastructure
Domain
Physical AI, 3D Computer Vision, Multimodal Learning
Deliverable
production ML models
Required skills
Deep learning (PyTorch/JAX), 3D computer vision, multi-view geometry, distributed training, large-scale data pipelines, evaluation methodology, GPU cluster operations
Preferred skills
PhD in 3D vision/embodied AI, web-scale video corpus curation, auto-labeling techniques, vision-language post-training, evaluation infrastructure (VLMEvalKit)
Technologies
PyTorch, JAX, Linux, Ray, Spark, Slurm, FSDP, CUDA
Responsibilities
Source and curate large-scale real-world image/video corpora for vision-language training; Build annotation and auto-labeling pipelines for 3D-grounded supervision; Own data quality end-to-end including semantic deduplication and quality scoring; Verify end-to-end model training pipelines and diagnose regressions; Build and operate 3D spatial evaluation suites and public benchmarks; Partner with research scientists to translate hypotheses into measurable capability; Operate large multi-node GPU clusters for data throughput optimization
Seniority
Senior, hands-on IC
