homescreen
image detection and nutrients display
Daily log screen
training model

NutriChoice

Overview

Many people struggle to understand what they are eating — especially when dining out.
Nutritional information is often missing, confusing, or inconvenient to check manually.
This can lead to overeating, unbalanced diets, or missed health goals.

NutriChoice is a mobile app built with React Native (Expo) that helps users make healthier food choices.
Simply point your camera at a dish, and the app uses an AI model to recognize the food and instantly display its nutritional values — including calories, proteins, fats, and carbohydrates — powered by a custom dataset.

Motivation

The idea came from a common problem: not knowing what exactly we eat, especially at restaurants or cafeterias.
I wanted to create something simple yet impactful — where AI bridges the gap between food recognition and informed dietary decisions.

How It Works

Frontend: Built with React Native (Expo) for a seamless cross-platform experience.
AI Model: A TensorFlow/Keras model (converted to TensorFlow.js) classifies food images into categories.
Dataset: Combined public food datasets with a custom dish.json mapping that links dishes to nutritional information.
Integration: Captured images from the camera/gallery are converted into tensors, processed by the model, and the top prediction is mapped to its nutrition profile.

Model Performance & Limitations

This is an early version of the model, trained on fewer than 400 images.
Current validation accuracy is around 64%, so misclassifications can occur.
Future updates will focus on:

Expanding the dataset for better generalization.
Improving accuracy through more training epochs and model tuning.
Optimizing performance for real-time mobile inference.

Key Learnings

Integrated a machine learning model into a mobile app using Expo + TensorFlow.js.
Understood the importance of curating clean, balanced datasets for classification tasks.
Designed a smooth user flow that translates AI predictions into actionable insights.

Challenges

Model Size & Performance: Mobile devices have limited resources, so optimizing the TensorFlow.js model was challenging.
Dataset Curation: Many public datasets lacked consistent nutrition info, so I manually created a mapping.
Testing & Compatibility: Ensuring consistent performance across Android devices with Expo.