Trashable

Inspiration

Growing up in a world where waste is everywhere, we realized that people actually want to recycle, but they are often confused by complex rules. Is this plastic bottle recyclable? Does this pizza box go in the paper bin? This "recycling anxiety" leads to contamination in our landfills. We wanted to build TRASHABLE to take the guesswork out of disposal and empower every individual to make a real impact on our planet with just a simple photo.

What it does

TRASHABLE is an AI-powered mobile assistant that identifies waste categories in real-time. A user simply snaps a photo of an item, and our app instantly classifies it into categories like Plastic, Glass, Paper, or Metal. Beyond just identifying the item, it provides immediate "Humanity Tips"—actionable advice such as "Rinse before recycling" or "Remove the plastic cap"—to ensure that the waste is processed correctly and doesn't end up polluting our oceans.

How we built it

We utilized a multi-layered tech stack to ensure the app is both fast and accessible:

App Framework: Built using Kodular for a responsive, user-friendly mobile interface.

The "Brain" (AI): We integrated a Quantized TensorFlow Lite (.tflite) model, which was pre-trained on a diverse dataset of garbage images to ensure high accuracy.

Edge Computing: By using the Personal Image Classifier (PIC) extension, the AI runs locally on the user's phone. This means the app works offline and respects user privacy.

Accessibility: Integrated Text-to-Speech to make the app inclusive for users with visual impairments.

Challenges we ran into

Our biggest hurdle was the "1GB Model Crisis." We initially found a massive dataset that was too large for a mobile device. We had to learn the hard way about Model Quantization—the process of shrinking an AI model from gigabytes down to a few megabytes without losing its "intelligence." We also struggled with integrating the PIC extension in Kodular, which required careful management of hidden WebViewer components to act as the AI’s engine.

Accomplishments that we're proud of

We are incredibly proud of achieving an offline-first AI. Most AI apps require a constant internet connection, but TRASHABLE can help a user even in a remote park or a basement. We also successfully managed to get the "Confidence Score" logic working, so the app doesn't just guess—it tells the user how sure it is!

What we learned

This hackathon was a crash course in Edge AI. We learned that building for humanity means building for constraints—limited data, limited battery, and limited storage. We also learned how to parse complex JSON-like lists in Kodular to extract meaningful data for our users. Most importantly, we learned that technology is most powerful when it’s used to solve small, everyday problems like "Which bin does this go in?"

What's next for TRASHABLE

We want to take TRASHABLE from a classifier to a community. Our roadmap includes:

Gamification: Adding "Eco-Points" for every successful scan to encourage sustainable habits.

Geo-Location: Integration with Google Maps to show users the nearest specialized recycling centers for e-waste or hazardous materials.

Brand Recognition: Training the model further to recognize specific brands and provide data back to companies on their packaging's lifecycle.

Built With

• android-camera-api
• block-based-coding
• google-text-to-speech
• kodular
• personal-image-classifier-extension
• quantized-neural-network
• tensorflow-lite
• webviewer