Lab / Playable
MDR Terminal
A browser game built in Unity in seven days with AI-assisted code: refine the scary numbers, like the Macrodata Refinement team in Severance.

Why I made it
I wanted to see how far a week and a lot of curiosity could go. Turns out: a whole refinement terminal.
- 01Shipping in seven days beats polishing for seventy.
- 02AI-assisted code got me to playable fast; game feel still took the hours.
- 03Unity → HTML5 means anyone can play it, no download.
Built with
AI: ChatGPT (wrote the code)
The tools are public. The tuning isn't. Ask me how →
Part of
What it is
MDR Terminal: The Refiners App is a small browser game inspired by the TV series Severance. You sit at a Macrodata Refinement terminal and work through a vast grid of numbers, looking for the irregular ones. Click and hold, sweep across a cluster that feels wrong, and release to refine it into a bin. Get it right and your bin fills and your file advances. Get it wrong and nothing happens at all, which is somehow worse.
It runs in the browser, with mouse controls for refining and arrow keys or WASD and the scroll wheel to pan and zoom the grid. It is a non-commercial fan project, published on itch.io under the studio name RISO.
How I built it
7 days, 17 to 24 March 2025
The game was built in Unity and exported as an HTML5 build, so it plays straight from the itch.io page with no download. Development ran for seven days, from 17 to 24 March 2025, and the page credits it as made with love, confusion and unease.
AI was part of the process from the start. ChatGPT wrote the code, and the work in Unity was implementing it, wiring it together and tuning it until it felt right. That kept the week focused on the feel of the grid, the selection gesture and the pacing of the bins and files rather than on boilerplate. The Lumon logo and design elements come from the Severance Wiki under a CC BY-SA 4.0 licence.
What I learned
A seven-day limit is a scope tool. Everything that did not serve the core loop of finding, selecting and refining was cut, and the game is better for it.
With AI writing much of the code, the bottleneck moved from building to deciding. The hours went into judgement: how a selection should feel under the cursor, how much feedback a wrong answer deserves, when a file should advance.
That is the same shift I bring to client pipelines. When the repetitive work is handled by tools, people get their time back for the decisions that actually need them.
More from the Lab
Next
