Source of Data & AI
Last updated: July 19, 2026 · Applies to the Palatify web app and mobile apps.
Palatify does not provide medical advice, clinical guidance, or treatment plans. Everything in the app is general nutrition information for educational and decision-support purposes. It is not intended to diagnose, treat, cure, or prevent any condition, and it never substitutes for a physician or registered dietitian. Palatify makes no disease-related claims and is not marketed for medical use.
Where every nutrition number comes from
Unlike most nutrition apps, Palatify attaches a source to every value it shows, at one of three levels:
Partner-verified — reviewed and confirmed by the restaurant itself through our partner tools; corrections are versioned and never silent.
Published — taken from the restaurant's own public nutrition disclosure (for example, chains that publish per-dish nutrition), with the retrieval date recorded.
Estimated (±) — computed by our engine from the public menu, grounded against federal reference data (USDA FoodData Central), and always displayed with an uncertainty label. An estimate is never presented as an exact fact.
Your personal targets are calculated with the Mifflin–St Jeor equation from the profile you provide — the same standard used across the fitness industry — and, once you have enough logged history, calibrated against your own observed energy balance. The app tells you when calibration is active and why.
How our AI works — and its limits
Three features use a large language / vision model: the nutrition coach, photo and description meal estimation, and reading a restaurant's menu into per-dish nutrition. All three run on Anthropic's Claude, server-side, through the DataRobot LLM gateway. Your chat text or meal photo, plus your targets if you're signed in, is sent to generate the response; we don't use it to train models and never for advertising.
The Fit Score is not generated by a model. It is a fixed formula — five weighted sub-scores over the nutrition values — which is why every ranking can show its working rather than asking you to trust it.
AI estimates are uncertain by nature, so Palatify treats them accordingly: photo estimates show a confidence level, every value is editable before it reaches your log, and your corrections are kept. If the model is unreachable, the app says so — it never invents a number to fill silence. The coach will not diagnose, prescribe, or advise on medication, and will point you to a clinician for medical questions.
Allergies — read this
Allergen and condition notices are generated from menu text and your profile. They are advisories, not guarantees: menu descriptions are incomplete and kitchens change. Always confirm allergens directly with the restaurant before ordering. When you order through Palatify, your flagged allergens are passed to the restaurant with the order — but that transmission also does not replace direct confirmation.
Reference data
Population-health context cited in the app comes from public federal sources. These are the primary ones, linked directly rather than paraphrased:
- CDC — Adult obesity facts (BRFSS, NCHS)
- USDA ERS — Food expenditure series
- USDA FoodData Central — nutrient reference
Restaurant names and brands remain the property of their owners; references are for identification only.
Corrections
Found a wrong number? Restaurants can correct their menus through the partner tools; diners can flag values in the app or email support@prosperiumars.com. Corrections are versioned with what changed and when — never applied silently.