AI agent development for businesses and clinics
I build AI agents that do a specific job inside an operation: answering and booking over WhatsApp, phone, Instagram, Messenger, email and web chat, or answering natural-language questions about the company’s data. I connect them to the real systems, such as a clinic’s calendar, and run them in production with tests, alerting and on-call. The most complete example is SuiteMedica’s booking agent, which works across those six channels.
What is included
- A customer-service and booking agent for WhatsApp, phone, Instagram, Messenger, email and web chat, which puts the appointment in the real calendar rather than in a separate spreadsheet.
- Business rules around the model. At SuiteMedica, the agent prioritises patients by their health insurance and their condition.
- Natural-language questions about the company’s data, like the AI module in CARS Control for sales, stock and finance, the one in ICV OS, or DataClinic’s chat about a clinic’s appointments.
- Automatic summaries of repetitive work, like the weekly surgical-quote summaries in MED360.
- Access to the models through an intermediate interface (OpenRouter in the published cases), so the model can be swapped without rebuilding the agent.
- Automated tests, monitoring and alerting from the first deployment.
How I work
- Before writing code I look at how the work arrives today: who answers the messages, at what time and in which system.
- The agent reads from and writes to the systems the operation already uses. If it cannot see the calendar, talking well is not worth much.
- Every change goes through continuous integration. At SuiteMedica that means build, type checks, lint, unit tests and end-to-end authorization tests between clinics.
- After launch I stay responsible, with on-call, alerting and monitoring.
Cases and products behind it
Frequently asked questions
Which channels can an AI agent cover?
Whichever ones your customers use. SuiteMedica’s agent answers over WhatsApp, phone, Instagram, Messenger, email and web chat with the same logic, and puts the appointment in the clinic’s calendar.
Which AI model do you use?
It depends on the case. I have worked with Claude, Gemini and GPT models. In the published projects the models are reached through OpenRouter, which lets you swap the model without rebuilding the agent.
Can the agent work with my current systems?
Yes, that is the point. The agent connects to the calendar, ERP or database you already use. If that system cannot be modified, the work is built on top of it, as with CARS Control.
Do you only work with clinics?
No. Most of the agents I have in production are in healthcare, but the approach works in other operations: CARS Control answers questions about an automotive group’s sales, stock and finance.