BinGo: Waste Management

Our app utilizes lightweight computer vision to help people identify different types of waste and lend advice in how to dispose them properly. We aim to achieve a cleaner, greener environment!

Comment

Inspiration

Every day millions of people misclassify their trash which overloads landfills and causes over $2 Billion dollars in annual damages in the United States alone and over 375 billion dollars in damages globally as of 2025. We were inspired to build BinGo as proper waste disposal is a substantial problem both economically and for the environment. Our project shows demonstrates that technology can be used to improve the environment.

What We Built

BinGo is a mobile application that uses on-device Ai to classify a picture that a user takes of trash they would like to dispose. The app then classify's the image from one of 12 categories:

  • battery
  • biological
  • brown-glass
  • cardboard
  • clothes
  • green-glass
  • metal
  • paper
  • plastic
  • shoes
  • trash
  • white-glass

From here the App uses a decision tree to classify the product as recycle, landfill, compost, hazardous waste and informs the user where they should dispose of the item with around 90% accuracy

Build

BinGo was build during the Hack for Humanity 2025 with a clear focus on speed, accuracy and a great overall user experience. The Build can be broken down into a few components:

Mobile App:

  • Developed Using React Native. The app features a responsive, clean interface that works with IOS and Android. Users can take a photo of their waste and view the results instantly

Local Ai Inference:

  • We created a custom trained model with over 15,000 images of trash and trained the model using mobilenetV2 (90% accuracy)

Challenges & Growth

  • Real time inference:
    • While initially building the model, our inference time was high (15-30 seconds) but with some fine tuning we were able to reduce the inference time to < 3 seconds.
  • Accuracy
    • Initially the model has an accuracy rate of 65% but after adding more images to the training set we were able to improve the accuracy. ## Try it out

Github Repo: https://github.com/Kieran-42/BinGo/tree/production-test

What we learned

How to create a Native React app utilizing machine learning.

What's next for BinGo

We would like to give it searching capabilities to quickly find local recycling guidelines for the user. Additionally, we would like to expand the current dataset to include more accurate waste identification.

Share this project:

Updates