Made a walkthrough video of zappy — a 42 team project. C++ server by PepeSegura and samuelcordero, Godot GUI by Tagamydev, Python AI by me.

The AI part: each client runs a custom behavior-tree that handles async TCP commands, partial vision, and no shared coordinate system. Coordination is solved through a queen protocol — drones use the broadcast message direction as a Marco Polo signal to regroup without ever sharing a map.

Watch on YouTube · Full write-up

Launched datacodemath — my own corner of the internet, built with Astro and deployed on Vercel. The publish endpoint means posts can come directly from a mobile app via API. The site itself was vibecoded with two AI instances working in parallel through a shared coordination file. The meta is not lost on me: the first thing I’m publishing is a site built by AI to publish things.

Made a video walking through the slither RL project — a Deep Q-Learning agent that learns to play Snake from scratch. The part I wanted to show: the agent has no idea it’s playing Snake. Game, Agent, and Interpreter are fully decoupled. Worth 8 minutes if you’re into RL or just want to see a snake eat itself for science.

Watch on YouTube

Finishing up leaffliction — a plant disease classifier built end-to-end in PyTorch. Six augmentation types, make-driven workflow, training metrics committed alongside the model. The interesting part wasn’t the model architecture — it was building the data pipeline and augmentation suite before touching any training code.

GitHub

Built an interactive fractal explorer in C — Mandelbrot, Julia, Cubic, and Newton sets. Implemented complex arithmetic from scratch because using <complex.h> felt like cheating for a project that’s fundamentally about the math. Newton fractals are the interesting ones: you’re watching a root-finding algorithm create structure.

GitHub