The Zetech AI Tribunal

A ruthless, cyberpunk-themed terminal interface powered by Gemini 2.5 Flash that acts as the supreme judge for the Zetech Prompt Battle, ranking and "roasting" student poetry submissions in real-time.

Comment

Inspiration

During our hackathon's "Prompt Battle," we wanted to create a judging system that matched the high energy of a live coding battle. We imagined: What if the judge was a cynical, hyper-intelligent system administrator from a dystopian future? Thus, the Zetech AI Tribunal was born.

What it does

The application is a Single Page Application (SPA) designed to look like a retro CRT monitor.

  • Ingestion: The event organizer inputs student names and their generated poems into the "Console."
  • Staging: Submissions accumulate in a visual "Staging Buffer," waiting for judgment.
  • The Protocol: When the "Execute Judgment Protocol" button is pressed, the app packages all submissions and sends them to the Google Gemini API.
  • The Verdict: The AI (acting under strict persona instructions) analyzes the poems based on Humor, Creativity, and References to Zetech University. It returns a ranked leaderboard complete with a "System Roast"—a sarcastic comment about each entry.

How we built it

  • Frontend : We used React (via Vite) and TypeScript to ensure type safety and fast performance.
  • Styling : Tailwind CSS was heavily utilized to create the "Green Screen" terminal aesthetic, including custom animations for glowing text, scanning lines, and blinking cursors.
  • The Brains : We integrated the Google GenAI SDK (@google/genai). We specifically used the Gemini 2.5 Flash model for its speed and ability to adhere to complex persona instructions (System Instructions). We engineered a prompt that forces the AI to output strictly structured JSON while maintaining a "funny but mean" personality.

Challenges we ran into

  • Prompt Engineering : Getting Gemini to consistently return valid JSON while simultaneously being creative and funny with its "roasts" took several iterations of prompt tuning.
  • The Aesthetic : simulating a realistic CRT monitor effect (scanlines and text glow) without making the text unreadable for the audience was a UI balance act.

Accomplishments that we're proud of

  • The "System Roast" feature is consistently hilarious and context-aware.
  • The UI feels immersive; using the app feels like hacking into a mainframe.
  • We successfully implemented the new gemini-2.5-flash model for near-instant results.

What's next for The Zetech AI Tribunal

  • Adding Text-to-Speech so the AI can read its own roasts out loud.
  • Adding a "Crowd Favorite" voting mechanism.
  • Persisting the leaderboard to a database for historical records.

Built With

Share this project:

Updates