Inspiration
The idea for EcoSort AI came from a simple situation experienced many times by most of the people- standing in front of a trash can and not knowing which bin an item belongs in. Even when people want to recycle correctly, they often end up guessing. I realized that the problem isn’t that people don’t care, it’s that waste sorting is confusing. A small mistake, like putting the wrong item in a recycling bin, can contaminate an entire batch and send it to landfill. I wanted to build something that could remove this confusion and help people make the right decision instantly. That’s how EcoSort AI was created- to turn uncertainty into clear, simple actions.
What it does
EcoSort AI is an iOS app that uses machine learning to identify waste in real time using the phone camera. Users can scan an item, and the app will instantly classify it as recyclable, biodegradable, or non-recyclable.
How we built it
I built EcoSort AI by combining machine learning with iOS development. For the machine learning model, I used Python and PyTorch to train a custom YOLOv9 model in Google Colab. I collected and prepared a dataset, split it into training, validation, and test sets, and trained the model over multiple epochs while tuning hyperparameters. The model was trained to classify three categories: recyclable, biodegradable, and non-recyclable waste. It achieved strong performance, with over 80% mAP@50 and accuracy up to 92% for biodegradable waste and 87% for recyclables. After training, I converted the model into Apple’s Core ML format and integrated it into an iOS app built with Swift, SwiftUI, and Apple’s Vision framework. This allows the app to run the model directly on the device and provide real-time results without needing the internet.
Challenges we ran into
One of the biggest challenges was making the model accurate enough for real-world use. Some waste items look very similar but belong to different categories, which made classification difficult. Another challenge was optimizing the model to run efficiently on a phone. I had to balance accuracy and performance so that the app could provide fast results without lag.
Accomplishments that we're proud of
I am really proud of making the product which is actually useful and helps the environment.
What we learned
I learnt how to train and test models and how to optimize it for mobile devices.
What's next for EcoSort AI: Machine Learning powered Waste Sorting App
I want to make an Android app to expand it to more users.
Log in or sign up for Devpost to join the conversation.