Senior 3D Spatial AI Engineer Grounding & Multimodal

NVIDIA; United States; On-site; USD 180,000 - 240,000; Full time

Job summary

NVIDIA Cosmos is hiring a Senior Deep Learning Engineer to own the 3D data engine and end-to-end training verification for spatial reasoning capabilities of 3D perception models. You will curate large-scale real-world image and video corpora and build end-to-end data pipelines with researchers from Cosmos Lab.

You will also own annotation, labeling pipelines, data quality, and the evaluation suite—validating model training, reproducibility, and benchmarking across CV-Bench, BLINK, and other

Qualifications

• MS or PhD in CS/EE/Robotics or equivalent.
• 12+ years of deep learning systems experience.
• 3D vision, multi-view geometry, or SLAM expertise.
• Vision-language models: data and evaluation pipelines.
• Large-scale multimodal data pipelines and versioning.
• Distributed training on multi-GPU, multi-node clusters.
• Strong communication and collaboration with researchers.

Responsibilities

• Own the 3D data engine for Cosmos spatial reasoning and curate large-scale real-world image/video corpora.
• Build annotation and auto-labeling pipelines for 3D-grounded supervision and related tasks.
• Own data quality end-to-end: coverage, faithfulness, completeness, correctness.
• Verify end-to-end model training: pre-training and supervised fine-tuning pipelines, reproducibility.
• Build and operate the 3D and spatial evaluation suite and public/in-house benchmarks.
• Partner with Cosmos Lab researchers to turn 3D hypotheses into dataset and ablation experiments.
• Operate on large multi-node GPU clusters to optimize throughput and prevent bottlenecks.
• Ship results into Cosmos releases and benchmarks.

Skills

3D vision

Multi-view geometry

Vision-language models

Large-scale data pipelines

Distributed training

Python Pytorch JAX

Education

MS/PhD in CS/EE/Robotics

Tools

PyTorch

JAX

Linux

Job description

NVIDIA Cosmos is hiring a Senior Deep Learning Engineer to own the 3D data engine and end-to-end training verification for spatial reasoning capabilities of 3D perception models. You will curate large-scale real-world image and video corpora and build end-to-end data pipelines with researchers from Cosmos Lab.

You will also own annotation, labeling pipelines, data quality, and the evaluation suite—validating model training, reproducibility, and benchmarking across CV-Bench, BLINK, and other