Want to know if this job is worth applying to?
San Francisco, California, United States
Onsite
Sieve is a multi-modal lab curating the world's highest-quality training datasets, spanning video, audio, images, text, and 3D. We combine exabyte-scale data infrastructure and novel multimodal understanding techniques that push the frontier of foundation models. Video alone makes up 80% of internet traffic, and across modalities, data has become the enabling medium powering creativity, communication, gaming, AR/VR, and robotics. Sieve exists to solve the biggest bottleneck in the growth of these applications: high-quality training data.
We partner with top AI labs and did $XXM last quarter alone, as a team of ~30 people. We also raised our Series A from Tier 1 firms such as Matrix Partners, Swift Ventures, Y Combinator, and AI Grant.
Sieve is one of the most capital-efficient teams in AI, with roughly 30 people serving the world's leading AI labs across every major data modality. You'll join early, own problems end-to-end, and watch your work ship directly into the models defining the frontier.
Physical AI runs on data that captures how the real world looks, moves, and fits together in 3D. As a perception-focused Machine Learning Engineer at Sieve, you'll build the systems that turn raw egocentric, robot, and sensor footage into the training data that robotics and world-model teams depend on.
You'll bring the computer vision instincts of someone who has shipped perception for robots, devices, or autonomous vehicles, and apply them to datasets at a scale few teams ever touch. One week you might recover camera pose and scene geometry across thousands of hours of first-person video. The next, you might build a hand and object tracking pipeline, or design a benchmark that tells a lab whether their data actually improves manipulation policies.
Build perception pipelines for depth, pose, tracking, segmentation, and 3D reconstruction over large video and multi-sensor datasets
Apply SLAM, visual odometry, and structure-from-motion to recover camera trajectories and scene geometry at scale
Fine-tune vision and vision-language models for embodied tasks like hand-object interaction, activity understanding, and scene understanding
Develop automated QA and filtering systems that catch calibration errors, sync drift, motion blur, and other data defects before they reach customers
Create benchmarks and evaluation frameworks that measure how much a dataset improves downstream robotics and world-model performance
Work directly with frontier labs and robotics teams to turn vague data requirements into production pipelines
Ship fast, measure results, and iterate based on how the data performs in the real world
3+ years building production perception or computer vision systems for robotics, AR/VR devices, autonomous vehicles, or similar
Hands-on experience with SLAM, visual-inertial odometry, multi-view geometry, or 3D reconstruction
Strong Python skills and fluency with PyTorch; C++ is a plus
Comfortable with multi-sensor data such as camera intrinsics and extrinsics, IMU, depth, and time synchronization
Experience training, fine-tuning, or deploying modern deep learning models for vision
Sharp intuition for evaluation, edge cases, and what makes data actually useful for training
Comfortable owning a problem from the customer conversation to a deployed pipeline
Excited about Physical AI and the data that will power it
In person at our SF HQ
Experience with neural scene representations such as NeRFs or Gaussian splatting
Background working with egocentric or teleoperation data
Publications or open-source work in perception, 3D vision, or robotics
All roles at Sieve require you to be onsite in San Francisco 5 days per week.
