WaterYouDoin

• 

  Extension pop-up

• 

  Asking a factual question

• 

  Asking a computational-heavy question

• 

  Submitting an unnecessary prompt

• 

  Submitting repeated prompts

Inspiration

AI tools are incredibly powerful, but we noticed developers often use them for things that don’t actually need AI, like definitions, quick lookups, or repetitive prompts. We wanted to encourage more intentional AI use without blocking or judging developers. Encouraging a more productive use of AI and avoid rabbit holes of intellectual doom.

What it does

WaterYourDoin is a browser extension that locally analyzes prompts before they’re sent to an AI assistant.

• Factual prompts are redirected to a non-AI search engine
  
• Low-value prompts are blocked instantly
  
• Reasoning-heavy prompts trigger a gentle nudge to think first
  

All decisions happen locally, and the extension visualizes avoided AI usage as water savings using a fun ice-cube mascot.

How we built it

We built a Chrome extension using TypeScript and Vite. Prompt classification runs entirely in the browser using lightweight NLP heuristics with an optional ML fallback. No backend, no cloud inference, no external AI calls. Stats are stored locally using Chrome storage and shown in a popup UI.

Challenges we ran into

Tuning the classifier to correctly separate factual lookups from reasoning prompts was tricky. We also had to carefully handle edge cases like short but meaningful prompts and duplicate submissions—all without slowing down the user experience.

Accomplishments that we're proud of

• Fully local prompt classification with zero backend
• Over 180 unit tests covering real developer prompts
• A clear, friendly UX that nudges instead of blocks
• A fun mascot that makes sustainability visible and memorable

What we learned

We learned that small design nudges can meaningfully change behavior without removing user freedom. We also learned how far simple heuristics and local ML can go when thoughtfully designed.

What's next for WaterYourDoin

Next, we want to support more sites, improve prompt classification over time, and add clearer insights into long-term AI usage patterns—while keeping everything private and local.

Built With

• javascript
• node.js
• typescript
• vite
• vitest