• Improved runtime of C++ Vulkan GPU based DO-178 compliant AI inference engine from 150ms to 70ms,

by improving host program, optimizing host device memory transfers and improving the GPU kernel

• Wrote Vulkan HLSL GPU Compute kernel to perform mean pooling operation in neural network, which

allowed the AI inference engine to successfully run the vgg16 and vgg18 neural networks

• Made presentations and gave talks internally in the company about Vulkan and GPU compute, focusing on

how to design Vulkan host applications and high performance high parallelism GPU kernels

• Created testing tools to measure kernel performance and correctness automatically for new iterations

• Wrote CUDA kernels for high efficiency C++ CUDA application for real time collision detection, using

multi-resolution and vectorization, was able to achieve fast, sub 10ms runtime

• Developed a variety of online & offline profiling tools, to analytically determine performance of algorithms

• Achieved 30% algorithm speedup by creating automated scripts to tune function parameters

• Primarily developed fixes for bugs reported in the display abstraction layer of AMD’s C++ graphics driver

• Investigated and analyzed bugs using winDbg, and created reproducible setups for the bugs

• Performed first 64-bit Linux port of CoreAVI’s DO-178 safety critical Vulkan SC2 AMD graphics driver

• Developed a Vulkan app in C to superimpose an image detection neural network’s output on the input image

• Designed and setup a GPU compute demo using MIVisonX that will be used at future trade shows