Graphics & systems
Ryukijano
My graphics and systems work includes photogrammetry in Unreal Engine, CUDA C++, and physics-informed neural networks. From September to December 2024, I worked with the Science Museum Group and Leeds HELIX XR on the Dalton Mills reconstruction.
Sound: Know Me (Project 1) on SoundCloud.
Projects
- Agentic structure-from-motion2026
An RL agent that calls geometric tools on hard structure-from-motion pairs. An implementation of Gabriele Berton's research idea, still in progress.
- CUDA Blackwell Labs2026
Twenty-two CUDA labs on a DGX Spark, from memory probes to hand-written FP4 tensor-core instructions.
- PCOS edge agent2026
A local-first router that decides which on-device model, memory store or cloud service answers a request, across Chrome, Android, iOS and a Pixel Watch.
- Dalton Mills VR Reconstruction2024
A VR reconstruction of Dalton Mills after the 2022 fire, for the Congruence Engine, built in Unreal Engine 5 with Leeds HELIX XR.
- AWS AI/ML Scholar2023
AWS AI/ML Scholar, 2022–2023. I trained PPO agents in DeepRacer and finished in the top 15% of the scholarship league.
- Physics-Informed Neural Networks—
CUDA notebooks for Navier–Stokes and Burgers-type equations, with physics-informed neural network experiments.
- B3tt3r: 3D Reconstruction—
Stereo reconstruction from image pairs, combining Mast3r and Spann3r.
- CUDA Kernel Development—
A practice log of CUDA C++ kernels and CUDA-Q / cuQuantum exercises.