DEV Community

Aggrey Paintsil Ishmeal
Aggrey Paintsil Ishmeal

Posted on

I Built an AI That Answers Every Question With HTTP 418 — RFC 2324 Compliant

April Fools Challenge Submission ☕️🤡

This is a submission for the DEV April Fools Challenge

What I Built

Meet DecisionBrew AI — the world's first Enterprise Teapot Intelligence Platform, powered by Google Gemini and RFC 2324.

You type any question. A life-changing question. A career question. A pineapple-on-pizza question. DecisionBrew AI runs it through a sophisticated 6-step AI reasoning pipeline — parsing your semantic intent layer, heating water to optimal brew temperature, cross-referencing the global tea-leaf database, running 847 parallel tannin models — and after all that computation, delivers its verdict:

418 — I'm a Teapot.

Every single time. Brew Confidence: 17%. Useful: No. Actionable: Absolutely not.

The "Consult a Real Human Instead" button also returns a 418. We are in a loop now. The tea has gone cold.

Demo

🫖 Live App: https://decision-brew-ai.vercel.app

Code

GitHub: https://github.com/aggreypaintsil168/decision-brew-ai

The entire app is a single index.html — no build step, no dependencies, no framework.

The core of the joke lives in the system prompt sent to Gemini: Gemini then invents tea-jargon analysis like "tannin gradient overflow", "steep confidence interval: collapsed", and "infusion entropy: critical" — all wrapped in a fake enterprise dashboard with animated reasoning steps, live metrics, and a consultation history log.

How I Built It

Stack:

  • Vanilla HTML/CSS/JS — zero dependencies, zero build step, one file
  • Google Gemini 2.0 Flash API — prompt-engineered as a sentient teapot
  • Vercel Serverless Function — to proxy the API key safely in production

The interesting engineering challenge was making the uselessness feel premium. The fake reasoning pipeline animates step-by-step with real elapsed times. The dashboard tracks metrics — brews today, average confidence (always ~17%), total 418s served, and a running consultation history. It looks like a serious product. It is not.

The system prompt was the most important part. Constraining Gemini to respond only in JSON, only with confidence between 12–23%, and always in the voice of a dramatically self-aware teapot took several iterations. Temperature set to 1.1 for maximum absurdity.

DecisionBrew AI is a love letter to that joke. Every single interaction ends in a 418. The entire product exists to serve that status code with maximum dramatic flair. HTCPCP IYKYK.

On the Google AI side: Gemini 2.0 Flash is not decoration here — it is the entire engine of the absurdity. Without the API, you just have a button that says "418". With it, you get a sentient teapot that tells you your career question has a "kettle alignment score of 0.003" and that "the quantum brew field destabilizes upon observation." That is the joke. Gemini makes it different and funnier every single time.


418 — I'm a teapot. This project solves zero real-world problems. Built with love, Earl Grey, and deep respect for useless internet standards.

Top comments (0)