← All open roles
ML Engineer, 4D Action Segmentation
Develop the action segmentation pipeline that converts continuous 4D hand-mocap streams into structured, training-ready tasks. The role spans model design, data tooling, and evaluation infrastructure.
What you'll do
- Develop action segmentation models that parse 4D point-cloud streams from the glove.
- Iterate across transformer, point-cloud, and hybrid architectures, characterizing strengths and failure modes of each.
- Build labeling and verification tools that close the loop between data collectors and the training dataset.
- Define evaluation suites that catch regressions before production.
- Partner with simulation, robotics, and product teams to align parsed output with downstream needs.
What we're looking for
- MS or PhD in CS, robotics, or related field, or equivalent applied experience.
- 3+ years working on sequence models, point clouds, or video understanding.
- Strong Python; fluency in PyTorch.
- Track record of taking models from research code to production-quality systems.
- Comfortable working at the boundary between learned models and structured pipelines.
Nice to have
- Published work in action segmentation, 4D understanding, or temporal models.
- Experience with HOI4D, EgoExo4D, or comparable datasets.
- Familiarity with point-cloud architectures (PointNet++, P4Transformer, PPTr).