Bin Guardian

Inspiration

Every year, roughly $6B in the U.S. is lost due to recycling contamination. We were motivated to build Bin Guardian to help reduce that waste by making recycling systems smarter, more efficient, and easier to use in public spaces.

What it does

Bin Guardian uses an image model to identify what type of bin a waste item belongs in whether it's recycling, compost, or landfill. The goal is to reduce contamination by correctly organizing waste items.

How we built it

We trained an image classification model using a mix of existing public datasets and our own custom-collected data. The model was optimized to run efficiently on low-power hardware, specifically a Raspberry Pi Zero 2 W, so it can be deployed affordably at scale.

Challenges we ran into

One of the biggest challenges was missing or incomplete data. Many waste categories didn’t have enough examples, especially in real-world conditions. To solve this, we created additional custom datasets to fill those gaps and improve model performance.

Accomplishments that we’re proud of

We successfully built a functional model that runs on a Raspberry Pi Zero 2 W. That means it’s lightweight, cost-effective, and practical for deployment in government buildings, schools, and public infrastructure.

What we learned

We learned how critical high-quality, diverse data is for real-world AI systems. We also gained experience balancing model performance with hardware limitations, especially when working with edge devices.

What’s next for Bin Guardian

Next, we want to build a full MVP system: a smart bin where users drop waste into a single input, and the system automatically sorts it into the correct category. Alongside that, we plan to expand our dataset, improve accuracy, and begin piloting in local and federal government environments.

Built With

• fastai
• linux
• mobilenetv3
• python
• pytorch