Systems programmer in specialization at 42 Nice, currently focused on AI & data science.
I spend most of my time close to the metal — memory management, process handling, network protocols. Understanding how things work at that level changes the way you write everything else.
Right now I'm finishing DSLR: implementing logistic regression and multi-class classification from scratch in Python — without sklearn — to understand what the abstractions actually hide before relying on them.
ft_transcendence — Real-time Pong with an AI opponent
Full-stack real-time Pong: WebSocket server in TypeScript, all game logic server-side, multiplayer sessions synchronized in real time. The AI opponent uses Q-leaning built from scratch — no library, just the algorithm, the state space, and a lot of tuning to make it actually competitive.
Minishell — Unix shell in C
The hardest part wasn't parsing commands. It was understanding how a real shell manages execution contexts, pipe chains, and signal handling simultaneously. Built a tokenizer, an AST, and process group management all from scratch.
ft_irc — IRC server in C++98
A real TCP server handling concurrent clients, channels, and operator permissions. C++98 is a raw environment — no lambdas, no smart pointers, no auto — that forces every design decision to be explicit.
Data science track — Linear regression → logistic regression
ft_linear_regression (complete): gradient descent from scratch on a car price dataset — normalization, training loop, cost curve, model persistence. No sklearn.
DSLR (in progress): reimplementing describe(), logistic regression, and multi-class classification from first principles. Same philosophy, harder math.
C / C++ — anything where control matters: memory, performance, syscalls.
Python — data pipelines, prototyping, ML experiments. Fast feedback loop.
TypeScript — real-time, event-driven systems.
Docker — because "works on my machine" isn't an argument.
Git — intentional commit messages, because history is documentation.
I'm looking for an internship where the engineering is the challenge — whether that's close to the metal or close to the model.