Inspiration
We wanted to create a way to inform people about sustainability issues and to teach them how to improve their habits in order to contribute to a more sustainable environment.
What it does
It gives the user a quiz on their habits and results from that quiz are inputted into a machine learning model to predict their CO2 emissions. Then it uses Claude's API to generate creative solutions to reduce their CO2 emissions based on the user's answers.
How we built it
We built EcoTrace with a Flask backend serving RESTful APIs and a dynamic JavaScript frontend featuring bubble morphism UI design. The Claude Sonnet 4 integration injects real-time context (current question, all previous answers, results breakdown) into every chatbot interaction, making it truly aware of the user's journey. We implemented an ML-informed calculator based on CatBoost model insights (R²=0.9907) trained on 10,000+ real carbon emission data points, providing scientifically accurate CO2 predictions for each user's lifestyle choices.
Challenges we ran into
Our biggest challenge was preventing Claude from "hallucinating" questions despite receiving full context—it kept asking "What question are you on?" even though we sent that data. We solved this through prompt engineering with strict rules and comprehensive context injection, while also pivoting from full CatBoost model deployment (Python 3.13 compatibility issues) to an ML-informed approach that extracted the model's coefficients for accurate calculations without heavy dependencies.
Accomplishments that we're proud of
We are proud of the end page which provides real and creative solutions that can lead to a more sustainable environment if followed correctly.
What we learned
We learned that prompt engineering and context structure are as critical as model selection for AI applications, and that making sustainability tools beautiful and friendly rather than intimidating is essential for driving real behavior change.
What's next for EcoTrace
We look to implement additional metrics like water usage and food waste. This entails gathering big data sets that relate individual habits to these metrics. We also look to include additional solutions in the form of alternative products that can be monetized through sponsors and affiliate links.
Log in or sign up for Devpost to join the conversation.