Trace · Los Angeles
Data & AIEngineeringRemoteFullTimeposted 3mo ago
Trace is building the data marketplace for physical AI.
Physical AI has the potential to transform how work gets done in the real world, from robotics to embodied systems that can see, move, and interact with their environment. But today, progress is constrained by a fundamental limitation: there is no scalable way to collect high-quality, real-world training data. Frontier robotics models are trained on orders of magnitude less data than language models because there is no equivalent of an "internet of robotics data."
Trace exists to change that.
We build the infrastructure that makes it possible to capture and transform real-world data from humans performing physical work, and turn it into training data for robotics systems, embodied AI models, and other AI systems that operate outside the browser and in the physical world.
If we succeed, we meaningfully accelerate the development of physical AI and expand what these systems can safely and reliably do in the world. Our platform is designed to support many data formats, capture workflows, and customer needs over time. What we capture today is only the starting point.
If you want to be an early hire at a company helping define how robots learn to work, keep reading.
A world-changing problem: Physical AI will reshape entire industries, but it cannot scale without real-world data. Trace is addressing one of the core constraints holding the field back.
Early but real traction: Active pilots with growing demand on both sides of the marketplace.
Experienced, tight-knit team: Ex-founders, PhDs, and operators with a track record of building and scaling together.
Real ownership: This is early. Your work will materially shape the product, systems, and direction of the company.
Foundational platform: We are building core infrastructure that enables many future products and use cases as physical AI evolves.
We are hiring a senior computer vision engineer to own the spatial perception layer of our data pipeline – the part of the system that turns raw, sensor-heavy data we capture into aligned, reliable representations the rest of the platform depends on.
This is load-bearing work. If calibration, localization, and trajectory recovery are unreliable, everything downstream – hand and pose annotation, object understanding, scene labeling, policy training – gets worse. Doing this well makes the entire output of Trace better, and our customers feel it immediately.
The work spans calibration, localization, mapping, pose estimation, and the failure modes that show up when you run perception systems against real-world data at scale. The specific sensor stack we capture on today will evolve over time, so we are looking for someone who is comfortable reasoning across software, sensors, and data quality rather than someone tied to a particular pipeline.
Own camera and multi-sensor calibration across our capture rigs, including intrinsics, extrinsics, and time synchronization
Build, evaluate, and improve SLAM, VIO, and mapping pipelines that recover aligned 6-DoF trajectories from real-world captures
Train and/or fine-tune models for pose estimation and semantic understanding of multi-modal data
Diagnose and fix the failures that actually show up in the field – drift, calibration drift, sensor misalignment, degraded tracking, weak reconstructions, noisy data
Define the ground-truth and benchmarking methodology we use to know whether the spatial layer is actually getting better
Decide where we need custom perception work versus where off-the-shelf components are good enough
Work closely with the rest of engineering and with Trace Labs (our applied research arm) to feed reliable spatial outputs into downstream annotation, evaluation, and product workflows
Strong experience in at least one of: SLAM, visual odometry, VIO, mapping, or localization
Hands-on work with camera calibration, sensor fusion, multi-sensor alignment, or state estimation
A track record of shipping perception systems on real hardware, in real-world environments – robotics, autonomy, AR/VR, drones, or other embodied / sensor-heavy systems
Comfort reasoning across software, sensors, calibration, and data quality, not just models in isolation
Pragmatism about when to use off-the-shelf components, when to build custom, and when to push a problem back to the sensor or capture side
High ownership, good judgment, and productive, thoughtful communication
Emotional maturity and a collaborative, grounded working style
Experience with reconstruction, SfM, pose graph optimization, or bundle adjustment
Work on multi-camera systems, LiDAR, or spatial computing
Prior exposure to large-scale data capture or sensor-heavy production pipelines