Research Scientist: 3D Hand Tracking & HOI

mecka; New York (NY); On-site; USD 190,000 - 260,000; Full time

Job summary

Mecka AI is seeking a Research Scientist to architect and train foundation models for 3D hand tracking and articulated pose estimation. You will design in-house architectures, leveraging a continuous stream of proprietary manipulation data to surpass public baselines, and own the hand-object perception loop within our data engine.

You will lead multi-view, multi-GPU training, push innovative losses for biomechanical realism, and tackle emergent perception challenges as hardware needs evolve.

Qualifications

• Deep expertise in Deep Learning, 3D vision, and articulated tracking/hand pose estimation.
• Proven experience training large-scale vision models from scratch.
• Strong understanding of hand models, inverse kinematics, and dense mesh estimation.
• Mastery of PyTorch and deep learning scaling frameworks.

Responsibilities

• Architect proprietary articulation models for 3D hand pose estimation and dense mesh recovery.
• Lead large-scale distributed training across multi-GPU clusters.
• Develop novel loss functions enforcing biomechanical constraints and temporal consistency.
• Build HOI (hand-object interaction) models for egocentric manipulation domains.
• Integrate tracking outputs into optimized pipelines informing robotic control policies.

Skills

Deep Learning

3D Computer Vision

Articulated Tracking / Hand Pose Est.

Large-scale model training

PyTorch

CUDA

Education

PhD in Computer Science / ML

Tools

PyTorch

CUDA

Multi-GPU Training

Job description

Mecka AI is seeking a Research Scientist to architect and train foundation models for 3D hand tracking and articulated pose estimation. You will design in-house architectures, leveraging a continuous stream of proprietary manipulation data to surpass public baselines, and own the hand-object perception loop within our data engine.

You will lead multi-view, multi-GPU training, push innovative losses for biomechanical realism, and tackle emergent perception challenges as hardware needs evolve.