-
-
Login page
Inspiration
With climate change and waste production increasing, it is ever more important for people to be aware of their surroundings and to become more educated in recycling. From young ones to seniors, everyone should be able to partake in the act of caring for the environment and the world around us.
What it does
EcoCycle Buddy allows users to chat and learn more about recycling and caring for the planet. With a text-based interface, users can send a message, and a recycling expert (fine-tuned LLM) will respond providing additional information. Additionally, users can take pictures or upload photos of their surroundings. A fine-tuned ResNet & YOLO model will detect what is in the image and inform the user (image2text). This model focuses on detecting garbage, recycling, etc. This information can be sent to the LLM where the user can ask more clarifying questions.
How we built it
The frontend technologies of the mobile app were React-Native, Expo Go (for development), and TypeScript.
The backend or server-side technologies were FireBase (user authentication) and AWS Lambda Function (to communicate with the LLM).
The Machine Learning technologies used were the ResNet50 model and YOLO. Specific training datasets of recycling text & images were found on Kaggle.
Challenges we ran into
An issue that I had when building the application was connecting the frontend app with the Lambda function hosted in AWS. I had not worked with AWS before, so the whole application was new to me. Additionally, when I typically develop ML, it is for fun and so I just host it locally. As a result, it was a new experience trying to host an ML model on the server. When connecting it, many 500s and 404 errors came my way, almost making me want to leave out the feature. However, in the end, it worked!
Accomplishments that we're proud of
The biggest thing that I'm proud of is finishing the app on time. Other than that, I am happy I was able to connect the ML models to the front end, specifically the LLM with the Lambda API.
What we learned
I learned a lot of things, mainly how to deal with server hosting. Hosting the ML models was a painful, yet great learning experience for me. It is something that I believe will be very useful as a skill in the future for me.
What's next for EcoCycle Buddy
There are many different directions in which EcoCycle Buddy can go. I am interested in two of these routes. One of which is to try to gamify the experience of learning and recycling. A leaderboard can be created where the person who takes a photo of the most items that they have recycled would be something I am interested in implementing. Such could also be to enhance learning and to create personalized quizzes based on the images that the user inputs along with the chats they have with the LLM. All of this I think would make recycling more fun and accessible to all, which aligns with the primary motive of the app.
Log in or sign up for Devpost to join the conversation.