Back to projects

GF FOOD PREDICTOR

VISIT LIVE SITE
GF FOOD PREDICTOR

ABOUT THIS PROJECT

GF Food Predictor is a fullstack machine learning web app that predicts what my girlfriend wants to eat, trained on her own eating history. Every meal is logged with its date, day of the week, whether it fell on a weekend, the meal type, her mood, and the budget, and that log becomes the training data for the model.

Under the hood it is a supervised learning classification problem. The logged features are encoded and fed into a Random Forest classifier, which learns patterns such as which dishes she reaches for at lunch on a weekday when she is happy and on a tight budget, and returns the most likely food for a new situation. A random forest suited this data well because it handles a mix of categorical and numeric features, copes with a small and noisy personal dataset, and is far less prone to overfitting than a single decision tree.

The app side is built with Next.js, Prisma, and PostgreSQL. It gives a clean food log where meals can be added, edited, and deleted, a tracked meals table, and a CSV export so the dataset can be pulled straight into the training pipeline. Keeping data collection, storage, and prediction in one product meant the model gets better simply by using the app every day.

The project taught me the full loop of applied machine learning end to end: designing what data to collect, building the product that collects it, choosing and training a model that fits the data, and serving predictions back to a real user.

← All projectsNext: CHATLY →

LET'S
TALK