This role pays more than
95%
of similar roles. Most pay
$202,475–$254,750
— the shaded band above.
At the midpoint, this role pays about
$290k
versus about
$229k
for comparable roles.
Based on 240 similar postings.
Employer
About Nvidia
Nvidia is a leading designer of graphics processing units (GPUs) and system-on-chip units, powering gaming, professional visualization, data centers, and artificial intelligence workloads. Industry: Semiconductors & AI Computing
Nvidia currently has
1150 open roles
on FindRole.
Listed pay typically runs
$184,000–$287,500
across 912 roles with salary data.
As a Senior Deep Learning Engineer on the Cosmo Engineering team, you will own the 3D data engine and end-to-end training verification for models providing 3D spatial reasoning and perception capabilities. You will build annotation and auto-labeling pipelines to produce 3D-grounded supervision, manage large-scale vision-language training data, and develop evaluation suites including benchmarks like CV-Bench and RoboSpatial. The role involves translating research hypotheses into experiments, managing multi-node GPU clusters, and optimizing data throughput for pre-training and supervised fine-tuning. You will utilize Python with PyTorch or JAX on Linux to solve problems in 3D computer vision, multi-view geometry, and structure-from-motion. Key technical requirements include experience with distributed training strategies like FSDP, as well as expertise in point cloud processing, depth estimation, and building large-scale multimodal data pipelines for robots and autonomous vehicles.
Source, curate, and filter large-scale real-world image and video corpora into vision-language training data for 3D spatial reasoning.
Build automated annotation and labeling pipelines to produce 3D-grounded supervision and multi-modal training data at scale.
Manage end-to-end data quality through semantic deduplication, automated scoring, and balanced sampling strategies for pre-training and fine-tuning.
Validate full model training pipelines by monitoring performance, diagnosing anomalies, and attributing capability changes to specific data decisions.
Build and operate 3D spatial evaluation suites and public benchmarks to measure model reasoning and geometry capabilities.
Translate research hypotheses from scientists into large-scale dataset experiments and measurable technical requirements for the Cosmos platform.
Optimize data throughput and sharding on multi-node GPU clusters to ensure high performance during long training runs.
What we're looking for
MS or PhD in Computer Science, Electrical/Computer Engineering, Robotics, or a related field, or equivalent experience.
12+ years of proven experience building deep learning systems in Python with PyTorch or JAX on Linux.
Deep expertise in 3D computer vision, multi-view geometry, structure-from-motion, SLAM, depth/camera pose estimation, point cloud processing, or 3D reconstruction.
Hands-on experience with vision-language models and building training data to improve visual grounding and reasoning quality.
Demonstrated experience building large-scale multimodal data pipelines including distributed video/image processing, deduplication, and automated quality metrics.
Experience running and validating large model training on multi-GPU, multi-node clusters using distributed training and sharding strategies.
Excellent written and verbal communication skills to partner with research scientists and translate research into engineering execution.
PhD and/or publications at CVPR, ICCV, ECCV, NeurIPS, ICLR, or CoRL in 3D vision, multimodal learning, or embodied AI (preferred).
Machine Learning
Computer Vision
Python
C++
PyTorch
JAX
OpenCV
Git
CUDA
3D Reconstruction
SLAM
structure-from-motion
NeRF
Gaussian Splatting
Neural Rendering
Multi-view Geometry
Agile Development
Computer Vision
Robotics
Python
C++
PyTorch
JAX
TensorFlow
SLAM
Structure-from-Motion
Optical Flow
Scene Flow
Vision-Language Models
Deep Learning
Sim-to-Real
GPU Clusters
3D Reconstruction
Multi-view Geometry