Egocentric AI Engineer

About Humyn Labs

At Humyn Labs, we believe the best AI is built on the best human judgment. We operate a global network of 1M+ verified experts who deliver high-quality, multimodal training datasets across domains — backed by reputation verification and multi-layer quality control. Our data isnt just collected; its evaluated, defended, and production-ready. Because before AI can be trusted, its training data must be.

About the Role

Youll be the engineering backbone of our egocentric dataset — building the pipelines, labeling tools, and quality infrastructure that our ML researchers and ops teams depend on. When the ML Researcher identifies a gap in what the model is learning, you build the data tooling to close it. When the ops team hits an annotation bottleneck or QA failure, you diagnose it and fix it at the root.

Your output is working software and robust pipelines. The quality of your data infrastructure directly determines the quality of our dataset — which determines what humanoid robots are capable of learning.

What Youll Build

• Egocentric video ingestion and processing pipelines — multi-camera sync, IMU/depth stream handling, calibration metadata, and MCAP/ROS format support.
• Labeling tooling: hand pose annotation interfaces, object tracking review tools, action segmentation editors, and 3D grounding visualizers.
• QA and validation systems — inter-annotator agreement scoring, label drift detection, and annotation audit dashboards that surface errors before they compound into training noise.
• Semi-automated annotation pipelines integrating ML models (pose estimation, open-vocabulary detection, tracking, VLMs) to reduce manual load and improve consistency.
• Dataset versioning and release infrastructure so researchers can run clean ablations and reproduce results reliably.

• Custom internal tooling in response to specific ops and ML team requirements — scoped from a brief, shipped as production-grade code.

How Youll Work

• Partner directly with the ML Researcher to understand data gaps and build tooling that closes them — translating research requirements into concrete engineering tasks.
• Work closely with ops and annotation leads whose throughput depends on your tools — triage their blockers, diagnose root causes, ship fixes.
• Occasionally join internal syncs with ops or BD stakeholders when a data problem needs an engineer to scope it properly — extract the real requirement, then go build.
• Feed patterns and recurring issues back into the data roadmap — youre a signal source, not just an executor.

Must Have

• 3+ years building data pipelines for vision or video datasets — egocentric or robotics strongly preferred.
• Hands-on experience with labeling infrastructure: annotation tooling, QA workflows, inter-annotator agreement, or dataset quality at scale.
• Strong Python engineering — production-quality code, efficient I/O, comfortable processing large-scale video and sensor data.
• Familiarity with egocentric datasets — Ego4D, EgoExo4D, EPIC-Kitchens — and their annotation schemas.
• Practical experience integrating vision ML models into data processing stacks (pose estimation, tracking, detection).

Looking to get Placed? Try our Placement Guarantee Plan

Get Hired
• Comfortable working across engineering and non-engineering stakeholders — able to extract a clear technical requirement from a fuzzy ops or research ask.

Nice to Have

• Experience with MCAP, ROS bags, or LeRobot data formats.
• Understanding of hand pose representations — MANO topology, 6DoF, axis-angle.
• Frontend experience (React or similar) to ship annotation and QA UIs independently.
• Prior experience at a data company, research lab, or robotics startup working across technical and non-technical teams.
• Contributions to dataset scaling studies (Ego4D, Open X-Embodiment, EgoScale) or open-source vision/robotics projects.

Who Youll Work With

• You report into engineering. Day-to-day you work alongside the ML Researcher and annotation leads.
• Your primary accountability is the quality and reliability of the data infrastructure you build.
• The cross-functional element — working with ops, ML research, and occasionally BD — makes you a better engineer:
• you hear problems at the source and build tools that actually solve them.