Member of Technical Staff, Computer Vision (Perception)
San Francisco • FullTime
Posted 6h ago
Job Location
San Francisco
Tech Stack
Remote Work Policy
On-site
Employment Type
FullTime
Categories
Machine Learning Engineer
About the job
Sieve is a multi-modal lab focused on curating high-quality training datasets across video, audio, images, text, and 3D, aiming to solve the bottleneck in the growth of AI applications. As a perception-focused Machine Learning Engineer, you will build systems that transform raw sensor footage into training data crucial for robotics and world-model teams. This role requires a strong background in computer vision, particularly in shipping perception systems for robots, devices, or autonomous vehicles, and applying these skills to massive datasets.
Responsibilities
- Build perception pipelines for depth, pose, tracking, segmentation, and 3D reconstruction using large video and multi-sensor datasets.
- Apply SLAM, visual odometry, and structure-from-motion techniques to recover camera trajectories and scene geometry at scale.
- Fine-tune vision and vision-language models for embodied tasks such as hand-object interaction, activity understanding, and scene understanding.
- Develop automated QA and filtering systems to identify and correct data defects like calibration errors, sync drift, and motion blur.
- Create benchmarks and evaluation frameworks to assess dataset impact on downstream robotics and world-model performance.
- Collaborate with frontier labs and robotics teams to translate data requirements into production pipelines.
- Ship solutions rapidly, measure their effectiveness, and iterate based on real-world data performance.
Requirements
- 3+ years of experience building production perception or computer vision systems for robotics, AR/VR devices, autonomous vehicles, or similar fields.
- Hands-on experience with SLAM, visual-inertial odometry, multi-view geometry, or 3D reconstruction.
- Strong Python programming skills and fluency with PyTorch; C++ experience is a plus.
- Comfort working with multi-sensor data, including camera intrinsics/extrinsics, IMU, depth, and time synchronization.
- Experience training, fine-tuning, or deploying modern deep learning models for computer vision.
- A keen intuition for evaluation, identifying edge cases, and understanding data utility for training.
- Ability to own a problem from customer interaction through to a deployed pipeline.
- Enthusiasm for Physical AI and the data that powers it.
- Experience with neural scene representations like NeRFs or Gaussian splatting (bonus).
- Background working with egocentric or teleoperation data (bonus).
- Publications or open-source contributions in perception, 3D vision, or robotics (bonus).