Mecka AI
Research Scientist- SLAM & VIO
$200K–$250K
New York · Hybrid · Full-time · Posted Jun 1, 2026
About the job
Build and train proprietary 3D spatial and egocentric optical-flow models, and integrate them into dense reconstruction and robotics perception pipelines.
What you'll do
- Design, implement, and train large-scale spatial models and optical-flow architectures using proprietary datasets.
- Develop models for dynamic-scene understanding, tracking, segmentation, and multimodal sensor fusion.
- Integrate model outputs with dense reconstruction pipelines, including 3D Gaussian Splatting and NeRFs.
What they ask for
- Deep expertise in deep learning, 3D computer vision, and temporal/video modeling.
- Proven experience training large-scale vision models from scratch.
- Strong understanding of modern feed-forward 3D networks and dense motion estimation.
- Mastery of PyTorch and deep learning scaling frameworks; experience curating massive image and video datasets.
Nice to have
- First-author publications in top-tier venues on relevant topics.
- Experience with egocentric video datasets or custom CUDA kernels for 3D operations.
Pay
$200K – $250K USD base salary
Tools
PyTorch, CUDA, 3D Gaussian Splatting, NeRFs, CVPR, ICCV, ECCV, NeurIPS
AI focus
Research and develop spatial-temporal perception models for embodied AI, including 3D geometry extraction, egocentric optical flow, dynamic-scene understanding, and dense reconstruction.
Model-written summary · the original posting is authoritative