The tech behind Bettie
Bettie looks simple on the surface, but it is powered by a small network of services. We wanted fast estimates and a clear way to refine them until they feel true.
When you snap a photo or type a meal, Bettie starts with a nutrition engine. We started experimenting with our own approach before we found the Tasty API, but their purpose-built nutrition AI consistently beat our baselines. They train on tens of millions of food images and fine-tune on USDA FoodData Central, FDA guidelines, and peer-reviewed nutrition research. They publish benchmark results and report 98.5% accuracy against laboratory standards, plus specialized computer vision for portion estimation and mixed-meal analysis. That level of detail is hard to match.
Tasty API
Tasty API handles image and text analysis with macros, micros, dietary flags, allergens, and preparation details.
Purpose-built nutrition models trained on 46M+ food images with benchmark accuracy against lab standards.
Explore Tasty APIOpen Food Facts
Open Food Facts adds packaged nutrition details, ingredients, and allergen data for branded foods and barcodes.
A global, community-built database of product nutrition facts, ingredients, allergens, and labels.
Visit Open Food FactsFrontier models
Frontier models power the chat layer and the OCR flow for receipts or menus so you can refine the estimate and correct anything that feels off.
Flexible routing across best-in-class models to keep chat and OCR fast and accurate.
No personally identifiable data is ever shared with these services.
The chat step matters because models can be too confident. We want you to be able to say “that was a larger portion” or “there was bread on the side” and see the numbers update immediately. The same loop applies when a receipt or menu needs a human correction.
Under the hood, we also generate embeddings to find similar meals in your history. That helps surface trends and gives you more context without extra work.
Our goal is to blend reliable nutrition data with a feedback loop you can trust. The tech should stay in the background so you can keep doing what you love, just with better information.