How we keep the AI safe (especially for clinics)

An AI that talks to patients must be held to a higher bar. Here is exactly what we do:

• A deterministic emergency gate. A separate safety layer — independent of the conversational AI — screens every clinic message for emergencies (severe pain, bleeding, swelling, trauma). When it fires, the conversation is escalated to your staff immediately with a summary. The AI cannot "decide" to skip this; it is a gate, not a suggestion. It also has a keyword fallback so it works even if the AI layer fails.
• Structured testing before every model change. We run scenario suites (emergencies, bookings, pricing questions) against any AI model before it ever serves your patients, and we keep instant rollback ready.
• Real facts, not guesses. The AI answers prices, hours, and services from data you control in the dashboard — injected as facts, not left to the model's imagination.
• Human handoff. Any patient can ask for a human at any time; your staff see the full conversation and take over. While a conversation is escalated, the AI stays silent.
• Your data is not training data. Conversations are processed via API to generate replies — API traffic is not used by the model provider to train models.