Member of Technical Staff, Computer Vision

Mecka•New York, NY
•$140,000 - $200,000

About The Position

Mecka AI is building data infrastructure for robotics and embodied AI, including global systems for data capture, labeling, and hardware-enabled workflows used by AI labs and robotics companies. The company works with robotics teams to connect real-world data, simulation, learning-based systems, and deployed hardware. The Member of Technical Staff, Computer Vision will develop models that interpret challenging footage of people interacting with the physical world and convert it into accurate 3D data for robotics training. The role focuses on computer vision specialties such as pose estimation, SLAM, and 3D reconstruction, with responsibility extending from model experimentation through reliable production deployment. The engineer will work across the perception stack, evaluate model failures on real data, and maintain shipped models and components.

Requirements

  • Strong machine learning fundamentals and comfort building and training models with Python and PyTorch.
  • Experience training large vision models from scratch, beyond running or fine-tuning existing models.
  • Deep experience in at least one computer vision specialty, such as pose estimation, SLAM, or 3D reconstruction.
  • Experience shipping a model or perception component to production and remaining responsible for it afterward.
  • Ability to make progress with noisy real-world data and adapt when priorities shift.

Nice To Haves

  • Experience with 3D hand and full-body tracking, mesh recovery, human-object or human-scene interaction, or estimating contact with objects and the ground.
  • Experience with learned ego-motion, optical flow, online calibration, or metric scale in SLAM, visual-inertial odometry (VIO), and calibration.
  • Experience with feed-forward or per-scene 3D reconstruction, Gaussian splatting, NeRF, or neural 3D object estimation from egocentric video.

Responsibilities

  • Train vision models from scratch on large multi-view, multi-sensor datasets, and develop new losses and architectures when existing methods are insufficient.
  • Address challenging footage involving motion blur, fast camera rotation, heavy occlusion, and partial views of hands, bodies, and objects.
  • Evaluate models on company video, measuring real-world error and failure rates as well as public benchmark results.
  • Optimize GPU performance, make results reproducible, integrate models into the data pipeline, and ship releases that can be rolled back.
  • Reproduce customer-reported issues, investigate their causes, and deliver fixes.
  • Prototype solutions as new perception problems arise.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service