Boiler Balance

Boiler Balance, powered by ChatGPT, revolutionizes personal nutrition tracking by offering intelligent, personalized feedback on daily dietary habits. Sensitive, kind feedback is only a click away.

Comment

Inspiration

As college students who have just stepped out of the nest, many of us found seemingly mundane aspects of daily life, such as making smart decisions regarding our meals and executing those decisions, challenging due to the abundance of independence.

When it comes to food diets, we felt semantic feedback from an LLM (large language model) would be better, as it could provide more detailed suggestions for the provided context compared to other simple numerical analyses. Our main focus to make it more "human", in a way that will impact people through words that numbers never can.

What it does

BoilerBalance utilizes the user's daily diet information to provide personalized semantic feedback via GPT-4.

Here are some of the features:

  • Users can create accounts and access their data with their login details anytime they wish.
  • Users can set up their profile via the initial survey, specifying details such as age, gender, height, weight, ethnicity and diet/health goals, which will all be considered in the GPT-4 feedback report.
  • A daily log will be prompted each day, asking about the user's diet information.
  • If there is a mistake made in the log, users can go to the 'daily review' page, delete data, and re-log the data.
  • Users can click "Get Feedback on Day" to receive the GPT-4 generated feedback and report.
  • Users can view and edit their past logs, by navigating through the calendar.

How we built it

Using React, we made a website that users could interact with and use. We implemented a user login system and user data storage system with MongoDB and Express.js. Using Flask, we were also able to utilize OpenAI's Python module to utilize GPT-4.

Challenges we ran into

One particular technical challenge was getting GPT-4 to output a response of a certain format that we wanted, since it would differ every iteration. To do this, we needed to curate an adequate prompt and fine-tune the hyperparameters of the GPT-4 API arguments. Additionally, because our team was composed of members of different skill levels, it was challenge to divide up the work and support everyone equally as we originally wanted to.

Accomplishments that we're proud of

We were able to implement a login system with persistent memory and user data, and utilize this data to prompt GPT-4 without much uncertainty. We also implemented a designed webpage with React and connected it to the database and the Python code.

What we learned

We were able to learn what features are important in a diet, and why semantics and context are so important in diet feedback. From a technical standpoint, we were able to learn how to properly utilize the GPT-4 API and how to use MongoDB and Express.js as a database for user data.

What's next for Boiler Balance

We are planning to implement more features the user can customize, such as dietary requirements, exercise, water intake, etc. Additionally, we hope to add weekly and monthly summaries to our LLM-based feedback options, along with visualization methods of progression and feedback.

Share this project:

Updates