Computer science student at Nordakademie in Germany, currently on exchange at the University of Cape Town for CSC2042S Machine Learning.
Three assignments into CSC2042S. A binary and one-vs-rest perceptron written from scratch on Simpsons-MNIST. A multinomial logistic regression for multilingual text classification. An MLP on Fashion-MNIST swept across learning rates, optimizers and dropout, ending at 88.65% validation accuracy with Adam at 0.001, 3.59 points over the baseline. The reports cover the ablations, the loss curves, and the classes the model keeps confusing.
| Project | What it is | Stack |
|---|---|---|
| Far From Home | Travel journal, live on GitHub Pages | Astro, TypeScript |
| CSC2042S Machine Learning | Perceptron, NLP classifier and MLP coursework, private while the course runs | Python, PyTorch, NumPy |
| Digital logic circuits | And, Or and Not gates wired into a circuit path for PdSE at Nordakademie | Java 25, Maven |
macOS and Linux (Arch and Ubuntu), with dotfiles managed by GNU Stow. Hyprland, zsh, Neovim, tmux, mise.