A bilingual patient assistant on the website and WhatsApp

Problem

The clinic wanted round-the-clock help that feels human rather than botty: welcome new patients, manage appointment changes, handle ten kinds of emergency, and answer more than twenty-five common questions — in French and English, on both the website and WhatsApp.

Approach

The flows that must not go wrong stay rule-based: new patient, modify or cancel, emergency, FAQ. On top of them, a language model reads what the patient actually wrote, works out which flow they need, and notices what is missing. Forget to give an email or a date of birth and the assistant asks for that one field instead of restarting the conversation.

On the front end, a widget in the clinic's own branding embeds anywhere on the site. WhatsApp runs through the Cloud API with webhooks, ready for the clinic's production number. Every completed request arrives at the secretariat as a structured email summary.

Outcome

Patients get a calm, branded assistant on whichever channel they already use, and the clinic team stops reconstructing half-finished requests from voicemail and email.

Stack

• NestJS with OpenAI for intent and entity extraction
• PostgreSQL for conversation logs and patient requests
• Next.js embeddable widget matched to the clinic's brand
• WhatsApp Cloud API and webhooks
• Structured email summaries over SMTP and SES