Spatial Computing Engineer

Schemata•San Francisco, CA

4d•Hybrid

About The Position

We are seeking a highly skilled Spatial Computing Engineer to join our team full‑time. You will play a foundational role in designing, building and optimizing the 3D‑scene‑understanding systems and multimodal AI pipelines that turn raw spatial data into actionable world‑models for next‑generation simulation and training applications. This is a high‑impact, cross‑functional role: you will work end‑to‑end from cutting‑edge research prototypes to production inference and performance profiling, ensuring our applications understand complex environments and respond in real time across diverse deployment targets.

Requirements

PhD or 4 + years equivalent depth in 3D computer vision, robotics perception, graphics‑ML or related field.
Strong Python with deep‑learning frameworks (PyTorch ); confident in CUDA or compute‑shader programming.
Hands‑on with 3D data types: point clouds, meshes, NeRF/LERF representations, SLAM or occupancy networks.
Solid grounding in linear algebra, geometry and numerical optimization.
Demonstrated ability to convert research into reliable, maintainable production code and services.
Experience profiling GPU workloads and scaling distributed training or real‑time inference pipelines.

Nice To Haves

Tier‑1 conference publications (CVPR, NeurIPS, SIGGRAPH) or open‑source contributions in 3D AI.
Large‑scale data‑engineering / MLOps experience for 3D pipelines.
Reinforcement‑learning or embodied‑AI background.
Defense, aerospace or other regulated‑industry experience; active or ability to obtain U.S. security clearance.

Responsibilities

Research and prototype state‑of‑the‑art methods for 3D reconstruction, segmentation, spatial reasoning and world‑model learning (e.g., NeRF/LERF‑style multimodal models).
Design data pipelines that convert heterogeneous 3D assets (CAD, photogrammetry, LiDAR, simulation output) into unified, queryable scene‑graphs and knowledge graphs.
Integrate multimodal foundation models (LLMs, VLMs) with spatial data to power diagnostics, step‑by‑step instruction and autonomous evaluation in virtual‑training scenarios.
Implement GPU‑accelerated training/inference, synthetic‑data generation and large‑scale evaluation workflows in cloud and edge environments.
Collaborate with graphics and product engineers to ship mission‑critical features that blend neural perception with real‑time rendering.
Profile and optimize performance across varied hardware configurations, balancing fidelity, latency and memory.
Publish internally, attend CVPR / NeurIPS / SIGGRAPH, and translate the latest research into production capabilities.