Deep expertise in 3D Computer Vision and Neural Rendering.
Proven experience with large-scale 3D vision models using PyTorch.
Experience with multi-terabyte datasets for training.
Ability to write custom CUDA kernels for 3D operations.
Strong publication record in top-tier conferences focusing on 3D deep learning.
Responsibilities
Design and train advanced models for dense 3D geometry extraction.
Reconstruct high-fidelity digital twins of real-world environments.
Optimize 3D scene representations for client-facing viewers.
Prototype new architectures to enhance perception capabilities.
Integrate reconstructed environments into existing perception pipelines.
Benefits
Access to proprietary spatial and temporal ground-truth data.
Opportunity to impact the future of embodied AI.
Collaborative environment with cutting-edge technology.
Full Job Description
The Role
Mecka AI is building the data infrastructure layer for robotics and embodied AI. While our existing perception division handles classical state estimation, this role is dedicated to bridging the simulation-to-reality gap through state-of-the-art spatial intelligence. We are hiring a Research Scientist to architect and train proprietary 3D reconstruction pipelines from scratch.
You will transform our continuous stream of high-quality, real-world multi-modal data into highly accurate, photorealistic digital environments using techniques such as 3D Gaussian Splatting and NeRFs. Beyond building the core architectures, you will integrate these reconstructed environments to improve other segments of our perception pipeline and optimize these spatial assets for seamless client delivery and interactive viewing. What You'll Work On
Advanced Neural Reconstruction: Design, train, and scale state-of-the-art models for dense 3D geometry extraction (e.g., Gaussian Splatting, NeRFs) across multi-GPU clusters to handle massive spatial datasets.
Sim-to-Real Integration: Reconstruct high-fidelity, physically accurate digital twins of real-world environments to provide rich spatial priors that directly improve and regularize downstream robotics pipelines, world models and other sim-to-real training.
Client Delivery & Viewer Optimization: Compress, optimize, and package massive 3D scene representations for high-framerate rendering in client-facing viewers, ensuring clients can interact with our data without sacrificing visual fidelity.
Emergent Perception R&D: Rapidly prototype new architectures to isolate dynamic actors from static environments, resolve algorithmic bottlenecks, and fuse novel hardware sensor integrations.
Who You Are
Deep expertise in 3D Computer Vision, Neural Rendering, multi-view geometry, and spatial modeling.
Proven experience training and scaling large-scale 3D vision models from scratch using PyTorch and multi-terabyte datasets.
Experience writing custom CUDA kernels to accelerate 3D operations, ray marching, or rasterization.
Strong Signals:
First-author publications in top-tier venues (CVPR, ICCV, ECCV) focusing on 3D deep learning, neural rendering, or spatial transformers.
Specific projects that include deploying 3D assets to WebGL, custom rendering engines, or optimizing splats for interactive viewing that are available publicly.
Why This Role?
The Data Advantage: Access proprietary spatial and temporal ground-truth data at a scale that most academic researchers only dream of.
High Impact: Your spatial models and environmental reconstructions will directly define how the next generation of embodied AI agents bridge the gap between simulation and the physical world.
A Note on Applying
Studies show women and candidates from underrepresented groups often only apply when they meet 100% of the listed qualifications, while others apply after meeting 60%. 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.
