Inspiration
As regular recyclers, my roommates and I often struggled to figure out what could actually be recycled. We wanted a simple, fast way to take the confusion out of sorting everyday waste, something that could help anyone make better recycling choices without second-guessing.
What it does
BinBuddy helps users identify where an item belongs (trash, recycling, or compost)just by snapping or uploading a photo.
It also gives quick, practical tips to recycle smarter and reduce contamination.
For example, if an image is classified as recyclable, BinBuddy can remind you to “Rinse before recycling” bridging the gap between awareness and action.
How we built it
Made BinBuddy using Streamlit for the app interface and OpenCLIP to analyze images. When a user uploads a photo, the model turns the image into a list of numbers (a feature vector). It then compares this image vector with the text vectors for each category (“recycle,” “trash,” “compost”). It chooses the category whose text vector is most similar to the image. That’s what the formula is saying, pick the label with the highest similarity score. Finally, the app shows the predicted category with clean CSS styling, making the tool clear, simple, and educational for users.
Challenges we ran into
- Getting OpenCLIP to work smoothly with Streamlit’s caching system.
- Handling dependency installations (PyTorch, torchvision, open-clip-torch).
- Fine-tuning the layout and color palette to look clean and stay centered across different screen sizes.
Accomplishments that we're proud of
- Building a working AI-powered waste sorter that feels intuitive and friendly.
- Customizing Streamlit’s interface to match a cohesive green theme.
- Creating clear, actionable recycling tips that help users learn as they go.
What we learned
We learned how to:
- Integrate a machine learning model into an interactive web app.
- Use caching and device optimization for faster inference.
- Balance technical performance with a seamless user experience.
- Apply design thinking to make sustainability feel approachable.
What's next for BinBuddy
Next, we plan to:
- Make recycling categories more detailed, identifying specific material types (e.g., plastic #1 PET, paperboard, metal cans).
- Add a location-based recycling center finder, allowing BinBuddy to detect the user’s location and suggest the nearest recycling facility automatically.
- Continue improving the model’s accuracy with a broader dataset and real-world testing.
- Eventually, turn BinBuddy into a mobile app for easier everyday use.
Log in or sign up for Devpost to join the conversation.