Smart Sorter

Inspiration

We wanted to make sustainable waste disposal. People often don’t know whether to recycle, compost, or trash an item, leading to contamination and inefficiency. Our idea was to automate waste sorting through AI-powered image recognition, while also educating users with short, clear recommendations on proper disposal methods.

What it does

Smart Sorter is a waste classification system that helps users instantly determine whether an item is recyclable, compostable, or trash.

Users upload a picture of an item (e.g. a bottle, wrapper, or container).

The image is analyzed by a TensorFlow Lite model we trained ourselves that classifies it into the correct waste category.

The app displays the prediction, confidence level, and a short recommendation on proper disposal.

Logged-in users (as protected by authentication) can view a history of their past scans, creating a personalized sustainability tracker.

How we built it

AI Model: We trained a custom image classifier using TensorFlow and converted it to TensorFlow Lite for faster inference.

Backend: Flask handles image uploads, model inference, and user authentication (via Flask-Login).

Frontend: Lightweight HTML/CSS interface styled consistently for usability, displaying uploaded images, predictions, and recommendations.

Database: SQLAlchemy stores user accounts and past scans to support tracking and future data visualization.

Challenges we ran into

Managing user authentication flow and weaving them with all the different aspects of our project. We also had alot of problems with TensorFlow version incompatibility across systems and model conversion from Keras to TensorFlow Lite (needed to handle unsupported arguments like groups in depthwise layers, which was very unexpected for us)

Accomplishments that we're proud of

Built a fully working AI-powered waste classifier with an interactive Flask web app.

Successfully integrated TensorFlow Lite inference for fast and efficient predictions.

Designed a user-friendly interface that makes sustainability accessible to everyone.

What we learned

How to deploy TensorFlow Lite models within Flask web servers, and how to implement secure authentication with Flask-Login. We also learned how to structure a full-stack web app to combine ML, backend logic, and frontend rendering effectively.

What's next for Smart Sorter

Improve model accuracy with expanded datasets and transfer learning. A mobile app version using TensorFlow Lite for real-time classification via camera input. Maybe gamification (eco-points, streaks) to motivate sustainable behavior as the user checks over time!