AI Transparency Notice

1. Purpose

This notice explains how PappAI uses artificial intelligence, which data are involved, and what governance controls are active. It is addressed to nutrition professionals using the platform and, where relevant, to their patients.

2. What the AI does

1. Drafts a proposed weekly diet plan based on the patient's clinical profile.
2. Generates operational insights on patient adherence and plan coherence.
3. Supports follow-up prioritisation by surfacing patients who may need attention.

3. What the AI does NOT do

1. It does not formulate medical diagnoses.
2. It does not perform automated clinical screening.
3. It does not prescribe therapeutic treatments.
4. It does not replace the professional's clinical judgement.

4. Model and fallback

1. Primary model: gemini-2.5-flash (Google Vertex AI).
2. Fallback: deterministic rule-based-fallback (local, no external API call).
3. Model pinning with CI guardrails prevents unauthorised model drift.

5. Data involved

PappAI applies a data minimisation principle: direct patient identifiers that are not strictly necessary for the nutritional task are excluded from AI payloads. The following are not sent to the AI model:

• Full name and surname.
• Tax identification number / national ID.
• Email address, phone number, or home address.

Only clinically relevant data (biometrics, dietary preferences, restrictions, goals) are included in AI requests.

6. Human-in-the-loop

1. All AI insights are created with an initial proposed status — they are never automatically applied.
2. Every clinical and operational decision remains with the nutrition professional.
3. Each AI output can be accepted, edited, or discarded by the professional.

7. Logging and reliability

Every AI call is logged with: endpoint, model used, outcome, fallback status, warnings, errors, and duration. Operational monitoring includes rate limiting, fallback/error-rate tracking, and automated alerts for anomalous behaviour.