ML Research Engineer, Robot Training Data

Telepath Robotics is the data layer for physical AI. We collect first-person (egocentric) video and manipulation data of real manual work at industrial and commercial sites across the Middle East and Central Asia, and turn it into training data for robotics and foundation-model teams.

We're hiring an ML Research Engineer to own data quality: the work that turns raw footage into data robot-learning teams can train on.

What you'll do

• Build pipelines that segment, label and structure egocentric video: task and action segmentation, hand-object interaction, temporal annotations.
• Define and automate quality metrics (coverage, diversity, label accuracy, capture defects) so every dataset ships with evidence of its quality.
• Run privacy processing at scale (face blurring, PII checks) while keeping provenance metadata intact.
• Package datasets in the formats robot-learning teams expect, with documentation and sample sets.
• Work with customer research teams to learn what makes data useful for their models, and feed that back into on-site capture protocols.
• Prototype with gripper (UMI-style) and multi-sensor data as capture expands beyond video.

What we're looking for

• 3+ years in ML engineering or applied research, with a focus on video or computer vision.
• Strong Python and PyTorch; hands-on experience with video models (action recognition, temporal segmentation, VLMs) and large-scale data processing.
• You care about data quality as much as model quality.
• Bonus: robot learning or imitation learning, egocentric datasets (Ego4D, EPIC-KITCHENS), annotation tooling.
• Comfortable in an early-stage team: high ownership, little process.

Why Telepath

• Foundational role: you'll shape how our data is built and measured.
• Real-world data from real worksites, not simulation.
• Fully remote, with some overlap with Dubai (GST) hours.