Job description

ABOUT MECKA AI Mecka AI is building the data infrastructure layer for robotics and embodied AI. We design and operate 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. We work closely with frontier robotics teams to bridge real-world data, simulation, learning-based systems, and deployed hardware. THE ROLE We're looking for a neural objects researcher to own the architecture for generating 3D objects from video. Given egocentric footage of someone manipulating an object, you build the models that recover its geometry — and, ultimately, a simulatable asset with the physical properties to drop into a physics simulator. You'll own the hardest, most open part of the objects effort: reconstructing objects that are small in frame, partly occluded by the hand, and sometimes deformable. You'll push the frontier, then hand results to the integrations researcher who takes them into the production pipeline. This is a research role for someone who wants an open, unsolved problem on real, large-scale data — not a fully-specified spec. WHAT YOU'LL DO - Neural 3D reconstruction. Models that recover object geometry from monocular / egocentric video — sequential and multi-view approaches robust to occlusion and low resolution. - Generative refinement. Use generative 3D to complete and clean up partial reconstructions while rejecting hallucinated geometry. - Toward simulatable objects. Move beyond geometry toward assets that carry physical properties and rigging, so a reconstructed object can be simulated, not just rendered. - Contact-aware reconstruction. Use hand contact as a signal to constrain object shape and pose where the hand touches the object. - Deformables. Take on the genuinely hard frontier — objects that change shape as they're handled. - Prototype & hand off. Build research prototypes and hand them to the integrations researcher for productization; publish where it makes sense. WHAT WE'RE LOOKING FOR - Strong 3D / neural-reconstruction research. Neural implicit or explicit 3D, 4D reconstruction, or generative 3D — you've worked at this frontier. - Deep-learning depth. You design and train models, and can read and reproduce frontier papers. - 3D geometry foundation. Structure-from-motion / multi-view geometry, meshes, and differentiable rendering. - Research taste. You can find the tractable decomposition of a hard, open problem and make steady progress on it. - Engineering to match. You prototype fast in clean Python / PyTorch and can get a research idea working on real data. STRONG PLUS - Generative 3D (diffusion / feed-forward 3D-generation families). - Differentiable rendering, NeRF, or Gaussian splatting. - 4D / dynamic-scene reconstruction from video. - Physics simulation and rigging — turning geometry into a simulatable asset. - Hand-object interaction; publications at top vision or graphics venues. TECH STACK - Python / PyTorch (primary) for model design and training. - 3D deep learning — neural implicit / explicit representations, differentiable rendering. - Generative-3D toolkits; mesh processing (Open3D, trimesh). - Physics simulators for the path from geometry to simulatable object. The exact stack matters less than depth in 3D deep learning and the judgment to make progress on an open problem. How this role fits — you own neural 3D object generation: reconstructing manipulated objects from monocular egocentric video, including the hard cases — heavy occlusion, low resolution, and deformable objects — and pushing toward objects that carry the physical properties needed to be simulated. This is an architecture-level research role that sets the direction the objects team builds on. WHAT SUCCESS LOOKS LIKE - Manipulated objects reconstruct from egocentric video at usable quality — including small, occluded, and deformable cases. - A credible path exists from interaction video to a simulatable 3D object. - Your research transfers into the pipeline through the integrations seat, not just into papers. - The objects pod builds on the architecture and direction you set. WHO THIS ROLE IS NOT FOR - Applied-only engineers who don't want open-ended research. - Researchers who need a fully-specified problem handed to them. - Anyone without a 3D and deep-learning foundation. WHY THIS ROLE AT MECKA - Define how a company turns everyday manipulated objects into simulatable 3D assets. - Own an open, high-impact research problem on real, large-scale egocentric data. - A clear path from research to product through a dedicated integrations seat. - Direct impact on the data that trains real robots. If you don't check every box above but believe you can do the job, we encourage you to apply — we're looking for capability and trajectory, not a perfect checklist match. We strive to create an inclusive work environment for all; we welcome applicants from all backgrounds and are committed to providing equal opportunities. Should you require any accommodations during the recruitment process, we will partner with you to meet your needs. We may use AI-assisted tools to help evaluate applications against the requirements of this role. These tools support our recruiting team in identifying qualified candidates but do not replace human judgment or limit which applications may be reviewed or considered.

Requirements

Must-haves

This is a research role for someone who wants an open, unsolved problem on real, large-scale data — not a fully-specified spec.
Neural 3D reconstruction. Models that recover object geometry from monocular / egocentric video — sequential and multi-view approaches robust to occlusion and low resolution.
Generative refinement. Use generative 3D to complete and clean up partial reconstructions while rejecting hallucinated geometry.
Toward simulatable objects. Move beyond geometry toward assets that carry physical properties and rigging, so a reconstructed object can be simulated, not just rendered.
Contact-aware reconstruction. Use hand contact as a signal to constrain object shape and pose where the hand touches the object.
Deformables. Take on the genuinely hard frontier — objects that change shape as they're handled.
Prototype & hand off. Build research prototypes and hand them to the integrations researcher for productization; publish where it makes sense.
Strong 3D / neural-reconstruction research. Neural implicit or explicit 3D, 4D reconstruction, or generative 3D — you've worked at this frontier.
Deep-learning depth. You design and train models, and can read and reproduce frontier papers.
3D geometry foundation. Structure-from-motion / multi-view geometry, meshes, and differentiable rendering.
Research taste. You can find the tractable decomposition of a hard, open problem and make steady progress on it.
Engineering to match. You prototype fast in clean Python / PyTorch and can get a research idea working on real data.
Generative 3D (diffusion / feed-forward 3D-generation families).
Differentiable rendering, NeRF, or Gaussian splatting.
4D / dynamic-scene reconstruction from video.
Physics simulation and rigging — turning geometry into a simulatable asset.
Hand-object interaction; publications at top vision or graphics venues.
Python / PyTorch (primary) for model design and training.
3D deep learning — neural implicit / explicit representations, differentiable rendering.
Generative-3D toolkits; mesh processing (Open3D, trimesh).
Physics simulators for the path from geometry to simulatable object.
The exact stack matters less than depth in 3D deep learning and the judgment to make progress on an open problem.
Manipulated objects reconstruct from egocentric video at usable quality — including small, occluded, and deformable cases.
A credible path exists from interaction video to a simulatable 3D object.

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aipython

Platform: ashby
Effort: quick

Benefits

Benefits detected in the listing.

Vision