Inspiration

We’ve all stood in front of a trash bin with a greasy pizza box or a plastic-lined coffee cup, wondering: "Does this actually go in the recycling?" Most people "wish-cycle"—they put items in the blue bin hoping they are recyclable, but this actually contaminates entire batches of waste. I built EcoScan to replace that hesitation with a clear, data-backed answer in under 5-8 seconds.

What it does

EcoScan is a mobile-first assistant that uses Gemini 3’s vision to identify household waste. It provides a simple, color-coded result: Blue (Recycle), Green (Compost), or Gray (Landfill). Beyond just a label, it explains why identifying specific materials like PET plastic or aluminum and gives a "Confidence Score" so the user knows they can trust the result.

How we built it

I utilized Google AI Studio’s "Build" environment to iterate on the application's logic and interface. The heavy lifting is done by Gemini 3 Flash, which I chose for its incredible multimodal speed. By engineering specific "System Instructions," I trained the model to analyze material density and contamination (like oil on paper) to provide accurate disposal advice without needing a massive pre-existing database of products.

Challenges we ran into

The biggest challenge was "The 10-Second Gap." Early versions took too long to process. I solved this by optimizing the model’s reasoning path and designing a "Progressive UI" that keeps the user engaged with status updates while the AI "thinks." I also worked hard to ensure the "Confidence Scores" were honest—if a photo is blurry, the AI actually tells you to verify manually.

Accomplishments that we're proud of

I’m proud of creating a tool that feels "invisible." You don't feel like you're talking to a complex AI; you feel like you're just getting a helpful tip from a friend. Converting raw AI power into a simple, three-color system for social good is what I find most satisfying.

What's next for EcoScan

The next step is Local Intelligence. Recycling rules change by city. I want to integrate geo-location so that the advice matches the specific rules of your local neighborhood.

Built With

Share this project:

Updates