Service
Implementation of AI for clinics and medtech
I use AI for routine work that takes hours. I call the rest what it is: hype.
The conversation often starts with “we need AI”. The actual problem turns out to be different: reception cannot keep up with messages, a quarterly report takes three days to compile by hand, or a doctor spends evenings answering the same patient questions. AI is useful for repetitive work, so I start by finding it. We choose the model and technology last.
Healthcare places particular demands on AI. It must not diagnose, invent facts or replace a doctor’s judgement. I therefore build systems where AI prepares and a person approves. The assistant handles administrative questions and helps patients book, while referring medical questions to a doctor. Reports are generated automatically but signed off by a person.
from $1,460· from the first conversation to launch. I prepare a detailed estimate after a short briefing.
Live examples:uzi.ru
Personally in touch at every stage. I value direct, personal communication. You work with me, without account managers or being passed to a colleague. We can discuss the terms, including flexibility and discounts. Get in touch and we will work it out.
What's included
An AI assistant that knows your business
flagshipThis is more than a button-based chatbot. The assistant uses your knowledge base, services, prices and schedule to answer in natural language and help people book. Try it at the top of this website or at the bottom of this page. In an ultrasound specialists’ community, the same kind of assistant helps doctors at their scanner. A QR code opens instructions for enabling features and configuring the device for a particular examination.
Self-generated reports
The data is already in your systems, but people still compile reports by hand. Regular reporting can be automated. A finished report arrives by email, as an Excel file or in Telegram. Work that took days can then take minutes.
- Bookings, enquiry sources and patients who attended
- Utilisation of doctors, rooms and equipment
- Ad spend and cost per enquiry by channel
- Revenue by service, doctor and branch
- Expenses by region and report to the head office
- Competitor prices and mentions of you
Routine work that runs automatically
Every company has repetitive work: checking enquiries, monitoring competitors’ prices, reminding patients about appointments or checking website availability. I move these tasks into scripts and AI so they no longer depend on someone remembering to do them.
Data the AI uses to find you
People increasingly ask assistants questions instead of using a search box. To name your business in an answer, AI needs to understand who you are, what you do and where you operate. I add structured data, build a knowledge base and check that answers about you are accurate.
Secure infrastructure and legal requirements
I deploy the system in a secure environment with role-based access and request logging. Data stays on servers in Russia as required by 152-FZ. Confidential patient information does not leave the protected environment; the model receives anonymised data. The infrastructure can be reviewed by the client’s security team and inspectors.
I calculate the payback before the start
First we calculate the time and money currently spent on the task. If AI implementation will not pay for itself, I say so. We can address the task with a written procedure or a simple script instead, at lower cost and with greater reliability.
How I work
- 01
We find the pain points
We examine where the team spends time and which questions recur every day. Staff usually identify these frustrations within half an hour of conversation.
- 02
We calculate the costs and return
We compare the task’s current time and cost with its expected cost after implementation, including running costs. That shows what to prioritise and what to leave alone.
- 03
A pilot for one task
I take one part of the workflow and get it working. A small pilot reveals answer quality and actual costs while mistakes are still inexpensive. There is no need to redesign the entire company around AI at the start.
- 04
Implementation and people
A system is useless if the team does not use it. I demonstrate it, gather feedback from daily users and adapt it to your actual workflow.
- 05
Observation and fine-tuning
I review logs and answer quality, then improve the wording and knowledge base. Models change quickly, so I design the surrounding system to let us replace a model without a rewrite.
Why can't a clinic just open ChatGPT?
The first thing a company that learns about AI does is create a shared chat and start putting work files there. This is not possible in medicine, and it is not a matter of convenience.
What must stay inside your organisation
Patient information, including the fact that someone sought medical care, is protected by medical confidentiality. Personal data cannot be sent freely to an overseas provider; specific storage and transfer requirements apply. An employee who pastes a table of patient names into a chat has already breached those requirements.
How is this solved?
We anonymise data before sending it to the model. The task often needs only figures and structure, rather than personal information. The database and all system outputs remain on servers in Russia, as required by personal data law.
How do we connect models?
We connect models through a Russian gateway. It provides access to several models and accepts Russian bank cards, with contracts and the accounting documents companies need. You do not need an overseas subscription, foreign bank card or legal entity abroad to get started.
What should the assistant say?
An AI assistant must clearly identify itself as AI. It must not pretend to be a human employee. Visitors often ask in their very first message and check the answer.
The review takes one conversation. It often reveals that half the tasks can be handled without using personal data at all.
What AI is already doing in medicine
In a clinic, AI first reduces the workload at reception. Repeated questions about prices, examination preparation, the address and available appointments can fill a receptionist’s working day. The assistant answers these questions around the clock and helps patients book. It refers medical questions to a doctor and does not diagnose. That boundary is intentional.
Medtech tasks are different and often have larger budgets. An equipment assistant tells a doctor how to enable a feature and configure a scanner for an examination. The doctor scans a QR code beside the device and asks in ordinary language instead of searching a manual. Head-office reports are generated from summary notes. Territory tracking, counterparty checks and enquiry reconciliation no longer depend on a manager finding time. Such systems can handle thousands of projects and hundreds of users at once.
Doctors scrutinise every answer. One invented fact can be enough to make them stop using the system. The assistant therefore answers only from your materials, whether equipment manuals, protocols, prices or schedules. If those materials do not contain the answer, it says so and refers the question to a person. This is the approach behind the assistant used by 20,000+ doctors.
Frequently asked questions
What does the price depend on?
It depends on the scope. An assistant using an existing knowledge base costs less than a system that draws data from spreadsheets and external services. Implementation starts at $1,460. Running costs are separate because every model response costs money. I estimate these before we start, rather than leaving you to discover them on the first bill.
How long does implementation take?
I do not promise a timescale before reviewing your data. We start with a pilot on one task to establish answer quality and real costs, then expand.
What if the AI lies to the patient?
The assistant therefore answers only from your knowledge base and refers medical questions to a doctor. I review conversation logs and refine the wording around real questions. Neither AI nor a human receptionist can be completely error-free, so we limit the consequences of a mistake.
Will our data leak?
We do not send patient data to overseas providers. That is explicitly prohibited, and we address it before starting work. The task usually needs only figures and structure, rather than personal information. See the section on data and the law above.
How is this different from ChatGPT for employees?
A general-purpose AI chat does not know your prices, schedules or services. It is not connected to your systems and does not answer your customers. It can help an employee draft text. We build a system that performs a specific job within your existing processes.
Who owns the system?
You do. I hand over the code, access credentials and data, and register keys to your company. If we stop working together, the system keeps running and another developer can maintain it.
We are medical tech, what should we automate?
We usually start with work compiled manually each quarter: reports, budget tracking, and checks on enquiries and sales territories. Another option is an equipment assistant for doctors and engineers that answers from your manuals. I have built both kinds of systems.
What if the model becomes outdated in a year?
It will become outdated, probably in less than a year. I therefore design the system so the model can be replaced without rewriting everything else. When a better or cheaper model becomes available, we connect it without changing how your staff work.
Can I see how it works before paying?
You are already looking at one. The assistant at the top of this website and at the bottom of this page is the same kind of system I install for clients, using my services and case studies as its knowledge base. Ask it a question.
Didn't find the answer?
Ask the assistant. It knows this service, the prices and my case studies, and replies immediately. It is the same kind of AI assistant I add to clients’ websites.
Related case studies
Ultrasound · community
UZI.RU
An educational platform for ultrasound specialists, with webinars, workshops and clinical case discussions. An AI assistant helps doctors use their scanner. They scan a QR code on the device and ask how to enable a feature or adjust settings for an examination.
Medtech AI reporting
Samsung Medison distributor
A marketing budget management system. It tracks spending and events and converts currencies at Central Bank rates. AI uses summary data to prepare a quarterly report for the head office in Korea and export it to Excel.
Shall we discuss your project?
Tell me about the task and I will suggest where to start. Not sure yet? Start with an initial audit for $300.
from $1,460· from the first conversation to launch