Software Engineer | C++ | CUDA | Python
Machine Learning, Computer Vision, 3D Vision, Image & Video Processing, Computer Graphics
ML systems, high-performance computing, distributed training, backend engineering, rendering and production pipelines.
Neural denoising for Monte Carlo-rendered sequences and video, focused on temporal stability and detail preservation.
Spatial and temporal models, Transformer architectures and attention experiments, motion compensation, custom CUDA kernels, FP16 inference and a standalone C++/CUDA inference runtime.
STeR - Stable Temporal Restoration improves temporal stability across video sequences by reducing frame-to-frame noise, flicker, ghosting and other visible temporal artefacts, with no per-shot tuning at inference.
Research: paper on ML-based temporal image and video restoration submitted for peer review.
Production spatiotemporal denoiser for Monte Carlo rendering in animation and VFX.
Combines spatial filtering with motion-compensated temporal processing to improve temporal stability while preserving fine image detail.
Deployed as a standalone application and Nuke plug-in across animated series and feature-film production.
Up to 7× reduction in visible Monte Carlo noise | ~30% lower rendering cost
Modular C++20 denoising framework designed for classical and neural image-processing pipelines.
Multi-layer and multi-frame image core, CPU parallelism, CUDA convolution, OpenEXR I/O and an extensible filter architecture.
Extensible Maya publishing pipeline for asset and shot production.
Versioned publishing, dependency tracking, scene validation and production metadata, with core logic decoupled from Maya and Qt for easier testing.
Core: C++ | CUDA | Python | CMake
ML: PyTorch | Transformers | Attention | Distributed Training | Triton | LLM Systems
Vision & Graphics: Computer Vision | 3D Vision | Image & Video Processing | OpenCV | OpenEXR
Backend & Infrastructure: FastAPI | REST APIs | Docker | Linux
Production: OpenUSD | Maya | Houdini | Nuke | RenderMan