EDY

Inspiration

Procurement teams at distributors like Arca Continental waste hours every day re-typing the same orders between a supplier portal and an internal tracking sheet. We wanted an agent that learns a process by watching, not by being programmed.

What it does

Edy is a Chrome extension with a floating assistant. It works in 3 phases:

1. Observes a person placing an order on a supplier portal (System A).
2. Learns the flow with Gemini — turning raw clicks into a reusable playbook.
3. Reproduces the order autonomously, even with brand-new data, and logs every order into Google Sheets + a live dashboard (System B).

How we built it

• A Chrome MV3 extension: a Shadow-DOM floating widget (UI), a content script (records actions + semantically finds elements on replay), and a service worker that orchestrates learning and execution.
• A Gemini backend on Vercel that converts a recording into a navigation playbook.
• A Google Apps Script + Sheets pipeline that auto-expands columns to fit whatever data the agent captures, feeding a live dashboard.

What we learned

That "learning a process" is mostly about separating the flow (the steps) from the data (the values) — that separation is what makes automation feel intelligent instead of scripted.

Built With

• chrome
• css
• gemini
• google-apps-script
• google-sheets
• html
• javascript
• manifest-v3
• shadow-dom
• vercel