Research Scientist - Computer Vision (3D Environment Reconstruction)

About 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.

Responsibilities

• 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.

Requirements

• 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.
• 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.