BreathalyzAir | Devpost

Inspiration

Inspired by the devastating Los Angeles and 2023 Quebec wildfires, which significantly impacted air quality in our area, we were reminded of the privilege of having access to accurate and regularly updated air quality index scores. This level of information is a luxury many communities around the world lack, even as they face prolonged and more severe air quality challenges.

What it does

We created BreathalyzAir to address this gap by empowering individuals in lower-income regions of the world with the ability to make informed decisions about their outdoor activities. For instance, if someone wants to determine whether it's safe to go for a morning walk, they can simply take a picture of the sky and upload it to BreathalyzAir. Using a neural network, the platform analyzes the image to estimate air quality and provides guidance on the safe duration to spend outside.

How we built it

We built the frontend of the project using HTML and CSS to design a user-friendly website and application interface. The backend leverages JavaScript for website functionality, while Python and Streamlit were used to develop and integrate the machine learning model that powers the application.

Challenges we ran into

Because we were new to model training, we spent a long time figuring out how to start the training, which took a while as we used a dataset of nearly 13,000 pictures. Unfortunately, the model is not entirely accurate or confident in its answers, resulting in incorrect assumptions some of the time. This issue can be attributed to the extensive data science required to study the pictures to determine air quality, which is out of the scope of the project. Another challenge we encountered was that we werent able to run our Streamlit application on anything but our local server we intend to solve this problem after DeltaHacks is over.

Accomplishments that we're proud of

This was the first neural network we had trained on a large data set, which was an important milestone as software developers and especially satisfying to see become functional. Overall, we are proud of our success in exploring technology which has the potential to have real-world impacts.

What we learned

We learned the principles of machine learning model development, user interface design, problem solving under time constraints, teamwork and collaboration, and most importantly, the potential of AI as a tool for social good.

What's next for BreathalyzAir

Presently, BreathalyzAir only works reliably for pictures taken in specific regions of India and Nepal, as the data set used to train the neural network consists only of pictures in those regions. However, with the help of crowdsourced images from around the world and improved data science, we can improve the neural network and increase the success rate of our app to help protect the respiratory health of people in all lower-income regions of the world. We also hope to launch a mobile app in the future to make this technology more user-friendly and accessible.