Discordia

Context

Discordia is the project I brought to my third-year exam: a web platform that works as a hub for Discord servers, built around a configuration generator driven by a language model. It was my testing ground for getting a website, an LLM and a bot to work together.

What I built

The core is the AI Builder: you describe in words the server you want, and the system generates the matching configuration. I ask the language model for structured JSON output — not free text to interpret, but data ready to use — so the result can be applied reliably.

Next to it is D.BOT, a bot that brings the features into Discord itself. The site and the bot share the same data, so what you configure on the site shows up in the server.

Technical choices

The platform is PHP with MariaDB on Apache, the stack I know best: it let me focus on the logic rather than the plumbing. For generation I use fast LLM providers (Groq, Cerebras) and always ask for structured JSON. The bot is Node.js with discord.js.

Forcing the model to produce valid JSON instead of prose is the choice that made the project solid: without it, every answer would have needed cleaning up by hand.

Status

Paused. The server is off for now; it comes back when I pick the project up again, with a few new features planned.