App interface

About the Project

Inspiration

I have always been interested in how food impacts health. Many people struggle to know whether their daily meals are balanced or not. This inspired me to build an AI-powered nutrition advisor that can classify meals as balanced or unbalanced using real nutritional data.

What it does

The project analyzes nutritional information such as calories, protein, fat, carbohydrates, and sugar. Based on defined thresholds, it predicts whether a given meal contributes to a balanced plate or not. This can help users become more aware of their diet and make healthier choices.

How we built it

The dataset was first uploaded and cleaned using Python and Pandas.
I used Scikit-learn to preprocess numeric features (scaling, handling missing values) and to train models.
Tried multiple models, and selected the one with best accuracy and ROC AUC.
Saved the trained model using Joblib.
Built a simple and interactive web app using Gradio and deployed it on Hugging Face Spaces.

Challenges we ran into

Handling missing nutritional values without losing too much data.
Defining clear rules for what counts as balanced vs. unbalanced.
Making the app lightweight so it runs smoothly on Hugging Face.

Accomplishments that I'm proud of

I successfully built and deployed a working AI nutrition app within hackathon time limits.
Learned how to go from dataset → ML model → deployment all on my own.
Created an app that can genuinely help people be more conscious about their diet.

What I learned

-How to preprocess real-world nutritional data. -How to train, evaluate, and save ML models in Scikit-learn. -How to use Gradio and Hugging Face for fast deployment. -How important it is to balance accuracy with simplicity in hackathon projects.