-
BU Terrier Trays Logo; Created by ChatGPT
Inspiration
Our group is interested in how we can better track our nutrition and meals on campus. Majority of our group members have a terrible diet, and we need something to tell us how to eat better on campus. In comes BU Terrier Trays, a helpful way to plan out your meals for the week based on certain guidelines including but not limited to allergies, halal, and cutting/maintaining/bulking.
What it does
Our project takes information from the user, scrapes online menu information from the BU websites for food information like macros, automatically create a 7-day meal plan based off of their caloric needs and other qualities, then present it to the user on our website.
How we built it
Brian thought of the idea, and we built out a plan of what the final product needs. We needed a scraper, regression models, data, and a web UI to pack it all together. With AI-assisted coding using ChatGPT and Claude, we used VSCode to write programs in Python (scrape, linear regression, etc). The linear regression model was needed in order to give the meal plans, like caloric and macronutrient intake. The website was created with react, tailwind, CSS, vite, and typescript with assistance from Claude. It was deployed on vercel.
Challenges we ran into
For creating the website, we had trouble having AI use our menu scraping and regression models. At first, they would not use our models at all. To fix this, we re-prompted until it worked. While developing our regression models, we encountered challenges with data leakage and overfitting. We learned that a high R^2 score doesn't mean a reliable model, especially when evaluated only on training data. We needed to prioritize generalization and proper model validation over training scores. We also had to convert everything into integers in regression, including categorical data, but categorical true or false was treated as 1 or 0.
Accomplishments that we're proud of
We are proud of our idea in particular; finding a gap on campus and pursuing how to improve it. It's fulfilling to see our idea come to life after hours of work. This accomplishment can inspire more apps that includes FitRec gym planning in a full-fledged nutrition tracker app. It's also the first hackathon for Aladin, Brian, and Erik, and we are all thrilled in the results!
What we learned
Even using AI to assist in code creation, it takes time to finalize an actual product. We learned that data plays a big role in apps that give recommendations, such as MyFitnessTracker. For cybersecurity, we learned how to properly identify cybersecurity risks and how to prevent them (now and in future prototypes), especially preventing API key leaks and user data leaks. Finally, we learned how to implement Figma, Dribble, and Mobbin into our front and back ends.
What's next for BU Terrier Trays
BU Terrier Trays hopes to become a full-fledged website that allows for users to login, track macros, plan their week out and more; possibly adding in school schedule guidelines that better help meal plans (Class on Bay State at Lunch? Go to Marciano's!). We also need to expand to include Fenway and Granby Commons. We could possibly include on-campus meal options with dining dollars. Think Terrier Transit, but for Terrier Trays! This could be a very helpful application for people looking to track nutrition at our university.
