Computational Physics, taught through stories.

Implement Llama 2 inference in C from scratch and then run a small LLM on a regular laptop. A nice excuse to look under the hood of the systems everyone keeps talking about, and to see firsthand why a 1 GB parrot says the things it says.

A student wanders the first floor of the Faculty of Science the week before exams, trying to ambush professors with last-minute questions. Unfortunately they are also a victim of doomscrolling: their eyes stay glued to TikTok, so they move in a correlated random walk. Stress builds up as time passes; finding a professor resets it, running out of time triggers a panicked flight from the building.

Simulate a small solar system with multiple planets orbiting a star — Verlet integration, gravitational forces, and a little data visualisation on the side. Integrating a system of coupled ODEs is always a bit tricky, and many students struggled in interesting ways.

A small 2D videogame, written in C against the SDL3 library, where a robot vacuum (a Rumba) tries to clean the first floor of the Faculty of Science. Random walk, collision detection, and just enough videogame physics that students could see their code in action.

The first assignment of the redesigned course — implement Euler and Verlet integration to simulate a harmonic oscillator, and compare their accuracy. In the exam, students were asked to add small modifications to their code, such as a slightly different potential.

Simulate a student who may have had a few too many drinks moving across a 2D grid, dodging buildings, and trying to make it to class on time. Random number generation, Monte Carlo, and a bit of collision detection.