DEV Community

Cover image for I Built an AI That Speaks Fluent LinkedIn (And It's Terrifying)
Sasanka Rath
Sasanka Rath

Posted on

I Built an AI That Speaks Fluent LinkedIn (And It's Terrifying)

April Fools Challenge Submission β˜•οΈπŸ€‘

This is a submission for the DEV April Fools Challenge


What I Built

Every year, millions of perfectly normal human thoughts are needlessly transformed
into impenetrable corporate speak. "I fixed a bug" becomes "I synergized the
codebase by disrupting legacy anti-patterns through agile remediation."

Meet BuzzwordAIβ„’ β€” the world's first bidirectional enterprise jargon translation
engine, built with the Google Gemini API. It translates plain English into maximum
corporate buzzword nonsense β€” and corporate buzzwords back into plain English.
It also generates a fake analytics panel after each translation: Buzzword Density,
Clarity Score, LinkedIn Readiness, and a smug AI observation.

There's a Jargon Level slider from 1 to 10:

1 = Slightly formal
3 = LinkedIn post
5 = McKinsey consultant
7 = Fortune 500 CEO email
10 = Sentient buzzword crisis
Enter fullscreen mode Exit fullscreen mode

And hidden in the corner is a πŸ«– teapot button. Click it. You're welcome.

We are solving a problem that AI helped create. With AI.


Demo

πŸ”— Live app β€” buzzword-ai.sasrath.com

Try these inputs:

Input Level What you'll get
I fixed a bug 5 A McKinsey-grade synergy statement
I need a nap 10 Existential corporate poetry
Let's leverage synergies to unlock net new pipeline — (Corporate→Human mode) "We need some sales leads."

Code

GitHub logo sasrath / buzzwordAI

πŸ”„ Because "I took a nap" deserves a quarterly business review. Gemini AI + HTTP 418 + zero sense. Built for the dev.to April Fools Challenge 2026.

buzzwordAI







How I Built It

I used this as an excuse to build something absurd with a production-grade stack.
Because overengineering April Fools projects is on-brand.

Stack:

Layer Technology
Framework Next.js 15 (App Router), React 19, TypeScript 5
AI Google Gemini 2.5 Flash-Lite (via server-side API proxy)
Styling Tailwind CSS 3, self-hosted (13 KB, no CDN)
Rate limiting Upstash Redis β€” sliding window 20 req/min per IP
Security Strict CSP, 8 security headers, HSTS
Deployment Vercel (serverless, 30 s timeout)

The interesting bits:

1. Gemini is server-side only.

The user's API key is sent in a request header to /api/gemini β€” Gemini is called
from Node.js with an AbortController 25-second timeout. The key is never logged
or stored server-side. The CSP enforces connect-src 'self' so the browser
literally cannot call the Gemini API directly.

2. Five distinct AI calls per translation.

One for the translation, one for analytics (buzzword density, clarity score, etc.),
and a few for edge cases. Each has its own system prompt with carefully engineered
output schemas so the analytics JSON doesn't end up as a philosophical essay.

3. Redis rate limiting with an in-memory fallback.

@upstash/ratelimit sliding window β€” 20 req/min per IP, max 3 concurrent. If the
Redis env vars aren't set (e.g. local dev), it falls back silently to in-memory
counters so development still works.

4. HTTP 418 Easter Egg.

The teapot button calls a special Gemini prompt that outputs a completely
straight-faced corporate memo that is secretly about being a teapot. It always
ends with "Short and stout." This is a tribute to RFC 2324 and Larry Masinter β€”
the author of Hyper Text Coffee Pot Control Protocol (HTCPCP).

5. Aurora UI with zero hydration mismatches.

The background has 4 animated CSS gradient blobs, 28 floating particle dots, and a
cursor-tracking 600px spotlight. The particles are generated deterministically from
index arithmetic (no Math.random()) to avoid React SSR/hydration mismatches.


Prize Category

Submitting for all three β€” because why not:

πŸ† Best Google AI Usage

Gemini is used for five distinct tasks per session: main translation, reverse
translation, AI analytics, the teapot easter egg, and the random-example seeder.
The prompts are carefully engineered to produce structured JSON from a generative
model consistently (temperature 1.0, strict output format instructions).

πŸ«– Best Ode to Larry Masinter

The HTTP 418 "I'm a Teapot" easter egg is the heart of the app. Hidden teapot button
β†’ modal quoting RFC 2324 β†’ Gemini writes a corporate memo that is secretly a teapot
confession β†’ output always ends with "Short and stout."

Larry Masinter deserves more credit for giving the internet its most beloved joke
status code.

❀️ Community Favorite

Because everyone has read a corporate email that needed translating. This is therapy.

Top comments (0)