How we use AI BETA

NutrientGraph is built with AI and checked with code. This page explains where AI is involved, where it is not, and what that means for you.

1. Where AI is used

AI models help build the dataset: they read primary sources and propose structured records for the matching of food records, the preparation state of foods, density references for drinks and the translation of food names. Two independent AI passes produce the records separately; disagreements are settled against the source text, and the reason is recorded. Where an AI made a judgment, the record says so, with the method and a confidence.

2. Where AI is not used

AI never types a number: code copies every value from the source file. Answers to your API calls are computed by deterministic code from the released dataset; no AI model runs when you call the API, and your requests are not sent to any AI provider or used to train models.

3. Checks

Every release passes code checks for whole classes of errors before it ships, and AI verifiers re-check samples against primary sources. The method and the current results are on the accuracy page. Expert review by licensed professionals is planned and not complete; we publish no accuracy percentage until it is.

4. Machine translation

Food names in French, German, Italian and Spanish are machine-translated unless the source publishes them (CIQUAL publishes French names). Each translated name is checked by a second, independent AI pass and labelled as machine-translated.

5. No decisions about people

The service does not profile people or make decisions about individuals. It returns reference data about products.

6. Report a problem

If a value looks wrong, write to support@nutrientgraph.com with the request and the response's meta.dataset_version. We check it against the cited source and publish corrections in the next release.

Version for the NutrientGraph beta. NutrientGraph is a trade name of ZAPPIT AI S.à r.l.