Norrisizer

A small, silly web app that generates absurd, hyperbolic "Chuck Norris Facts" using Google's Gemini API — and quietly personalizes them to your taste over time.

Built as a weekend project for Google AI Dojo Season 2.

What it does

• Generates a one-liner Chuck Norris-style joke, optionally themed around a topic you type in.
• You rate each joke (Love it / Too much! / Needs more / Not for me).
• Under the hood, an LLM call turns your ratings into a running bullet-point list of your taste, which is folded back into the system prompt for future jokes — no fine-tuning, just prompt-based personalization that evolves per session.
• Jokes you loved get collected into a small "best of" list the model uses as style reference (without copying them verbatim).
• A belt-ranking system (white → black) tracks how many jokes you've rated.
• Bilingual UI (Japanese / English).
• A couple of easter eggs (try mentioning a certain 80s pop song as your topic).

Stack

• Backend: Flask + Gunicorn
• Model: Gemma (via the Gemini API), with an optional unlock code to switch to Gemini
• Frontend: a single templated HTML page, no framework
• Deploy: Dockerfile included, designed to run on Cloud Run (reads PORT from env)

Running locally

pip install -r requirements.txt
export GEMINI_API_KEY=your-key-here
python app.py

Without GEMINI_API_KEY set, the app falls back to a couple of canned jokes so the UI still works.

Environment variables

Variable	Purpose
GEMINI_API_KEY	Gemini API key. Required for live joke generation.
UNLOCK_CODE	Optional code that lets a user switch from Gemma to Gemini.
PORT	Port to listen on (defaults to 8080).

Disclaimer

This is a joke generator in the classic "Chuck Norris Facts" internet-meme genre — exaggerated, absurd, and not intended to mock or harm anyone. It doesn't take itself seriously, and neither should you.