Streaming Point Cloud Visualizer
A Vulkan viewer that streams a city-scale, 13-billion-point scan within a 4 GB GPU budget.
View project
Piotr Rybicki
I design high-performance software for GPU computing, robotics, scientific applications, and large-scale visualization. I turn complex algorithms into reliable, practical tools.
Selected work
A Vulkan viewer that streams a city-scale, 13-billion-point scan within a 4 GB GPU budget.
View projectHardware-accelerated LiDAR and radar simulation for repeatable robotics perception testing.
View projectA responsive scientific tool for aligning scans, annotating neurons, and inspecting fluorescence traces.
View projectA CUDA solver that turns 3D geometry into a repeatable virtual flow test bench across multiple GPUs.
View projectBuilding GPU projects, including a viewer that renders a 180 GB, 13-billion-point scan at 120 FPS within a 4 GB GPU budget.
Implemented and validated low-level C++ features, including L1/L2 cache intrinsics, for a next-generation AI accelerator before silicon.
Led the architecture, API, GPU execution model, and performance of LiDAR simulation libraries sustaining 500M rays/s and peaking at 1.9B rays/s.
Reduced per-camera CPU use in the RealSense stack from 111% to 13%, enabling 16 cameras on one PC at about 150 ms capture latency.
Built Python and Qt tools with neuroscientists to annotate and analyze two-photon microscopy data.
Worked on CUDA driver internals and developed memory-copy tooling that helped achieve up to 4× speedups in selected cases.
Delivered a multi-GPU D3Q19 lattice Boltzmann solver, networking protocol, and real-time visualization client.
Built Gazebo simulations, integrated neural perception, and coauthored research on robust robotic grasping.
Parallelized protein-chain evolution on the GPU, reaching about 100× the CPU implementation's performance.
Got data that's outgrown your tools? That's my kind of problem.
Have a system to build, or want to talk shop over coffee in Warsaw? I'd be glad to hear from you.