Egor Khromov

3D Printing Compensation (ownership) Designed DAG architecture with API, validation & memory management Novel loss function + JAX with decomposed 3D convolutions (12x speedup) Automated calibration, cutting model prep time by 50% QC & Diagnostics Automated QC catching 99%+ of print issues Diagnostic tool used by 2 R&D teams to trace defects 3D Deep Learning Minkowski sparse conv + ViT backbone for multiple tasks PointNet++ for manufacturing Infrastructure ClearML, Git Flow; mentored 3 domain experts

- Collaborated with clinicians to define quality criteria for treatment plans and built a recommender-style model predicting doctor satisfaction with proposed plans - Co-invented a multimodal fusion approach combining 2D X-ray and 3D intraoral scans to extract more accurate root position data, awarded a patent - Evaluated Transformer and Graph Neural Network (GNN) architectures for Finite Element Analysis (FEA) surrogate modeling, informing team investment decisions

- Ran feasibility study on 4D sparse convolutions (Minkowski Engine) for spatio-temporal dental data - Built point cloud, mesh, and voxel preprocessing pipelines using Open3D, 3D Slicer, and Trimesh - Benchmarked PointNet++, Point Cloud Transformer, and sparse convolution approaches for 3D dental scan analysis