About
I’m Melvyn. I build things across machine learning, interaction, and graphics — and I care most about making hard problems feel simple, and about tools that give people back their time.
I like building from first principles — a NumPy autodiff framework, a physically-based ray tracer — and I’m always learning something new. A lot of it ends up here. English, French, Chinese.
Selected Projects
-
A deep-learning framework built from scratch, verified against PyTorch
Gradients and losses match PyTorch numerically. Enough of a framework to train real models, with every line accounted for.
-
A physically-based ray tracer that scales to O(log n) intersections
Intersection cost drops to O(log n), turning scenes that crawled into ones that render in a reasonable time.
-
Decoding neural signals into text with a CNN-GRU model
An end-to-end signal → phoneme → text pipeline, evaluated on edit distance against ground truth.
Skills
- Languages
- C++ Python JavaScript HTML/CSS SQL GLSL
- Domains
- HCI Computer Graphics Computer Vision Multimodal Deep Learning
- Tools
- PyTorch OpenGL ROS 2 CMake Linux Git