Executive summary
PappAI is designed as professional nutritional and organisational support software: it does not formulate diagnoses, does not perform automated clinical screening, and does not prescribe therapies. This document sets out the technical and legal controls that support this classification.
Declared intended use
- Operational support to the nutritionist in managing patient care pathways.
- Improvement of patient adherence and care continuity.
- Support for the creation and updating of nutritional plans.
Architectural controls
- Model governance: runtime pin on
gemini-2.5-flash. - Controlled local fallback:
rule-based-fallback. - Mandatory regulatory notice on all AI outputs.
- Data minimisation on AI payloads.
- Full audit trail on all AI operations.
- Rate limiting, alerting, and operational monitoring.
Non-SaMD classification rationale
- No therapeutic automation based on diagnosis.
- No automatic diagnosis: AI outputs are informational insights only.
- Mandatory human-in-the-loop on every clinical decision.
- Non-diagnostic framing enforced in prompts, outputs, and UI.
Residual risks and mitigations
- Semantic drift toward clinical claims: lexical guardrails and regression tests.
- Improper use by the professional: mandatory documentation and disclaimers.
- Uncontrolled model changes: version pinning and blocking CI checks.