Gianone/smartplate

SmartPlate – AI Nutrition Assistant

Photograph your meal and get instant nutritional analysis with evidence-based health advice from WHO, DGE, and Harvard guidelines.

How it works

1. 1.📷 Computer Vision — A Vision Transformer (ViT) fine-tuned on Food-101 classifies the dish (20 classes, 96.46% accuracy)
2. 2.🔢 ML Numeric — Logistic Regression classifies the dish as healthy/medium/unhealthy based on nutritional values (100% test accuracy)
3. 3.💬 NLP RAG — OpenAI gpt-4o-mini generates evidence-based advice from a vector store (ChromaDB) of WHO/DGE/Harvard nutrition guidelines

Supported food classes

Healthy: caesarsalad, greeksalad, edamame, misosoup, grilledsalmon Medium: sushi, sashimi, spaghettibolognese, padthai, chickencurry, omelette, pancakes, ramen Unhealthy: pizza, hamburger, frenchfries, donuts, cheesecake, icecream, chocolatecake

Project Repository

GitHub: https://github.com/Gianone-byte/smartplate

Built as a semester project for the ZHAW "KI-Anwendungen" Module (FS 2026).