Experience

Built and modified custom CUDA kernels for live video processing on Jugo’s immersive meetings platform, which renders real-time 3D meeting environments in the browser on an NVIDIA and Unreal Engine stack. Worked across the NVIDIA DeepStream video stack, adapting a game-engine render pipeline to live AI video compute and low-latency streaming under a hard per-frame budget. Profiled and optimised the compute and memory bound stages for throughput at scale, from one-to-one calls up to town-hall audiences, using Nsight to drive occupancy, launch configuration and memory-coalescing changes. Some details remain confidential; scope and results available on request.

Delivered the GPU-resident render pipeline for Union One, a tactical FPS: particle systems and related effects computed on the GPU rather than the CPU, with Vulkan driving rendering and the surrounding engine systems. Owned the subsystems where frame budget and memory layout were the binding constraints, profiling and optimising against a fixed per-frame time budget.