Projects
GPU-Accelerated B-Rep Healing for NVIDIA Omniverse
- Key Achievements: Integrated a GPU-based iterative solver (\AMGx{}) to reduce complexity from O(N³) to O(N).
- Impact: Enables rapid and robust healing of complex 3D geometry in real-time environments.
- Links: NVIDIA Omniverse
Skills: C++, CUDA, Nsight Systems/Compute, FEM, PDE Solvers, NURBS
Scalable High-Resolution PDE Solver for 3D Printing Simulations
- Key Achievements: Scaled to 1000+ processors, handled 500K DOFs and 3000 time steps in under 6 hours.
- Impact: Allows detailed thermal and stress analyses for voxel-based, time-varying geometry in additive manufacturing.
- Links: Bunny Video, Moai Video, Paper
Skills: C/C++, MPI, HPC, FEM, Adaptive Meshing, Python, Paraview, GDB
GPU-Accelerated Collision Analysis of Voxel Geometry
- Key Achievements: Achieved 300K CPU and 1500K GPU speedup over traditional methods.
- Impact: Enables interactive, real-time collision detection in complex 3D scenes (e.g., autonomous vehicles).
- Links: Paper
Skills: CUDA, C/C++, Nsight Compute, Unreal Engine, Point Clouds, Voxels, Memory Optimization
Bayesian Optimization for High-Throughput TEM Image Analysis
- Key Achievements: Used graph-based Bayesian optimization to reduce human intervention and improve accuracy by 30%.
- Impact: Facilitates near real-time image segmentation and parameter tuning for advanced materials analysis.
- Links: Paper
Skills: Python, Bayesian Optimization, Graph Algorithms, Image Processing, HPC
CUDA-Accelerated X-Ray Imaging and Deep Learning for Defect Detection
- Key Achievements: Developed a CUDA-accelerated X-ray simulator (4000x faster than CPU) and trained a U-Net for near-zero false positives.
- Impact: Improves manufacturing quality control and ensures robust defect detection.
- Links: Paper
Skills: CUDA, C/C++, GDB, Nsight VScode, Python, Deep Learning (U-Net), Physics Simulation
