- Website
- CRM
- AI chatbot
- Patient intake
- Medical document upload
- Doctor teleconsultation
- Hospital database
- Treatment-cost database
- Quotation system
- Visa workflow
- Travel coordination
- Patient dashboard
- WhatsApp communication
- Payment
- Basic accounting
1 Like
Don’t build this as one agent. Use one n8n workflow per job, plus one AI Agent node that calls the other workflows as tools.
Intake: a Form Trigger with a File field for the medical documents, or a Webhook from your site. Save the patient to your database with a Postgres node. Put the quotation, visa and travel steps each in their own workflow that starts with an Execute Workflow Trigger. Attach those to the AI Agent with the Call n8n Workflow Tool, so the chatbot can look up costs and build quotes. For messages, use the WhatsApp Trigger and the WhatsApp Business Cloud node. For payments, a Stripe Trigger on checkout.session.completed updates the patient record. Keep the dashboard, video calls and accounting in dedicated apps. n8n only moves the data between them.
Where does your hospital and treatment-cost data live today?
1 Like
I build exactly this stack for a real medical clinic (intake → documents → WhatsApp → quotations), so let me add the parts that only show up once you have live patients in the system:
**Start with the money path first.** Intake form → document upload → WhatsApp confirmation → quotation. That one pipeline proves the model to your team in 2–3 weeks, and everything else (visa workflow, travel coordination, dashboard) branches off it as separate workflows.
**Medical documents need a preflight step before any AI touches them.** Wrong file, blurry scan, wrong patient, 40-page PDF that’s actually 3 patients mixed together — I run validation first (file type, size, page count), then LLM extraction with confidence scores, and anything below threshold goes to a human review queue instead of silently mangling data. This one step kills most of the failure cases people blame the AI for.
**WhatsApp layer:** template messages for the transactional stuff (confirmations, quote delivery, payment links) — those are reliable and pre-approved. Route free-form patient questions to an AI agent with a strict toolset: look up treatment costs, build a quote, book a consult. The agent coordinates; it never gives clinical advice. That boundary matters both for safety and for how confident your staff feel handing it the phone number.
**Payments:** Stripe checkout per quotation, webhook updates the patient record in Postgres and triggers the visa/travel workflows automatically. One source of truth in the database, every workflow reads from it.
The teleconsultation, dashboard and accounting pieces belong in dedicated apps — n8n moves the data between them, as blessoftware said. The nervous system, not the whole body.
Happy to sketch the intake → quote pipeline with you in more detail. What does your patient intake look like today — paper forms, a portal, or WhatsApp screenshots?
Portfolio: alaneisenberg.tech | GitHub: github.com/eisen0x
1 Like
Something I’ve been thinking about for this kind of stack: state tracking and preflight validation feel like two pieces that are easy to skip early but hard to add later.
A status column per patient (intake → docs_received → quote_sent → visa_started → paid) gives every workflow a single place to check where things are — makes the AI Agent’s routing way more reliable than having it infer from context.
For medical document uploads, a validation step that checks file type, size, and routes low-confidence AI extracts to a human queue seems like cheap insurance against bad data propagating downstream.
Also curious what others think about isolating quotation logic in its own workflow — that way pricing changes don’t ripple into intake or visa flows. And a single webhook from whatever payment provider handles checkout could update the patient record and trigger downstream coordination together.
1 Like