SortWise

Inspiration

We wanted to build something with a direct positive impact, so we looked to our daily lives and the pressing issue of sustainability. One thing stood out: many people want to recycle but aren’t sure what belongs in which bin. This confusion leads to contamination and waste. We realized that with machine learning, we could make it much easier. Our goal was to empower anyone, regardless of age or ability, to make environmentally conscious decisions right from their phone or computer.

What it does

SortWise is a web application that uses image classification to identify what type of trash you’re looking at and tells you how to dispose of it correctly. Users upload a photo of an item, like a plastic container or coffee cup, and our trained model predicts the category (e.g., recyclable, compost, landfill) and provides disposal guidance. It’s simple, accessible, and designed to remove the guesswork from recycling.

How we built it

We built the frontend using React for a clean, responsive experience. The backend is powered by FastAPI and runs a custom-trained TensorFlow image classification model. We trained the model ourselves from scratch using a public trash image dataset over the course of 3 hours and 50 epochs. The classified result is passed back to the user via a REST API. Styling and layout were customized to be screen reader–friendly and easily navigable, especially for those with visual or cognitive impairments.

Challenges we ran into

Training the model took a lot of time and coordination. We encountered performance limits and device overheating, but stayed committed to getting the model up and running. On the frontend, we had to solve layout issues and ensure a smooth experience across different devices.

Accomplishments that we’re proud of

We successfully trained and deployed our own custom model, built a responsive web interface, and created a full classification pipeline from upload to result. We’re especially proud that we built this from scratch and not with pre-made AI wrappers. The result feels intuitive, fast (even on a limited backend), and helpful for users of all abilities.

What we learned

We learned how to train and deploy TensorFlow models, how to design an accessible user interface, and how to coordinate backend and frontend work under tight deadlines. We also gained a deeper appreciation for edge cases in UX, like ensuring users get clear feedback during wait times.

What’s next for SortWise

We plan to integrate location-based recommendations (e.g., nearby recycling centers), support more item types, and allow users to submit unknown items to help improve the model. We also want to make this tool more accessible in schools and communities, helping everyone become a little more sustainable, one snap at a time.

Built With

• fastapi
• javascript
• jupyter
• keras
• python
• react
• render
• tensorflow
• vercel
• vite