SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Mecka AI is building the data infrastructure layer for robotics and embodied AI. The company designs and operates global systems for data capture, data labeling, and hardware-enabled workflows used by leading AI labs and robotics companies to train and validate humanoid and embodied AI systems.
You will architect and train proprietary foundation models from scratch focused on 3D hand tracking and articulated pose estimation. Your core mandate is twofold: building in-house equivalents to cutting-edge 3D hand and mesh recovery architectures, and developing highly robust interaction models tailored for the heavily occluded, chaotic domain of egocentric manipulation. You will serve as a lead problem-solver for emergent perception challenges as hardware and downstream robotics needs evolve.
Key responsibilities include:
**Architecting Proprietary Articulation Models:** Design, implement, and train state-of-the-art networks for 3D hand pose estimation, dense mesh recovery, and kinematic tracking. Scale multi-view and temporal ML architectures across multi-GPU clusters to handle massive, multi-modal datasets. Develop novel loss functions that enforce biomechanical constraints, temporal smoothness, and physical plausibility.
**Egocentric Hand-Object Interaction:** Build and train custom architectures capable of handling extreme motion blur, severe self-occlusion, and rapid rotations inherent in first-person object manipulation. Track objects through complex grasps, segment tools from hands, and map contact points and forces to provide rich regularization for downstream action-conditioned robotics models.
**Emergent Perception R&D:** Rapidly prototype and deploy new models for tasks spanning tactile-visual fusion, fine-grained action segmentation, and novel hardware sensor integrations. Pivot to resolve sudden algorithmic bottlenecks in the data engine, adapting the latest research to unblock new product capabilities.
**Dense Contact & Physics-Aware Tracking:** Connect foundational tracking model outputs into highly optimized pipelines that reason about physical contact surfaces and object affordances, bridging human video data and robotic control policies.
You will have access to a massive, continuous stream of high-quality, proprietary ground-truth manipulation data captured by Mecka's infrastructure, enabling you to train networks that surpass current public baselines and own the complete hand-object perception loop for the data engine.
**Requirements:**
- Deep expertise in Deep Learning, 3D Computer Vision, and specifically Articulated Tracking / Hand Pose Estimation
- Proven experience training large-scale vision models from scratch, not just running inference or fine-tuning existing checkpoints
- Strong theoretical and practical understanding of parametric hand models, inverse kinematics, and dense mesh estimation
- Mastery of PyTorch and deep learning scaling frameworks
- Experience handling and curating massive, multi-terabyte image and video datasets for training
- Comfortable operating in a fast-paced environment where priorities can shift rapidly
**Strong Signals:**
- First-author publications in top-tier venues (CVPR, ICCV, ECCV, NeurIPS) focusing on 3D hand tracking, hand-object interaction (HOI), dexterous manipulation, or human mesh recovery
- Specific experience working with massive egocentric manipulation datasets (e.g., Ego4D, Ego-Exo4D, DexYCB, Epic-Kitchens) and solving the unique optimization challenges they present
- Experience writing custom CUDA kernels to accelerate 3D operations, differentiable rendering of meshes, or collision/contact computation
