Speech becomes documentation.
The data stays in your practice. MedScribe listens to the consultation and produces a structured SOAP note entirely offline, in the RAM of a device on your desk.
AI in the clinic and in research
Start from the basics with a trainer. Practise on examples from everyday work.
- Dates
- October 2026
- Length
- 2 evenings × 2:30 h
- Format
- online, on our training platform
What you will learn
- 01Anonymise recordsUsing synthetic visit notes
- 02Write precise promptsClaude · ChatGPT · Gemini
- 03Verify resultsCatching errors and fake sources
- 04Work with literatureOutlines and draft texts
Learn to work with AI in healthcare.
Start with the free S0 demo or go straight to S1. Over two online evenings you practise writing prompts in Claude, ChatGPT and Gemini, checking results, anonymisation and working with literature. We also cover the AI Act and the SynthID watermark. The course is for physicians and researchers, including those new to AI. You work on our training platform, with fictional material and a trainer on hand.
Data: recognise the risk
Removing a name is not enough when combined details can identify the patient. In S1 you practise anonymisation on synthetic visit notes, distinguish it from pseudonymisation and learn to recognise when data needs an assessment before it is shared.
Output: check the source
A fluent answer can contain invented facts and publications that do not exist. You compare the output with the source, check the citations and decide whether the text can be used. We also cover the SynthID watermark: its absence does not prove a human wrote the text, and its presence says nothing about whether the text is true.
For institutions
Before you enrol.
Is it for me if I don't know AI?
Yes. S1 starts with the basics and a platform warm-up. No programming or prior S0 attendance is required. The S0 demo lets you try the approach beforehand.
Online or on-site? How long is it?
S1, S3 and S3.5 run online. S1 and the Publication workshop (S3.5) are two evenings of 2:30 h, Vibecoding (S3) two evenings of 2:15 h. S2 Extension runs on-site over Saturday and Sunday, or online over three evenings. Groups are small, so there is time for individual questions.
Do I have to start with S1?
S1 is the starting point and the basis for the other courses. S2 requires S1. The specialist tracks, Vibecoding and the Publication workshop, assume you already know the basics. If you are not sure where to start, choose the free S0 demo.
What hardware and accounts do I need?
For S0 and S1 you need a computer with an up-to-date browser and internet access. We use our training platform, and login instructions arrive before the session. You do not need to buy any AI subscription beforehand. Requirements for other courses are confirmed at enrolment.
Do we practise on patient data?
No. S0 and S1 use fictional materials only. We practise removing or generalising identifying details and assessing re-identification risk. This exercise does not authorise sending real medical records to any AI tool. Do not include patient data in enrolment forms or emails.
What does the S1 certificate confirm?
After passing the S1 test you receive a completion certificate listing the programme. It can document the development of AI literacy, but it is not a certificate of an institution's compliance with the AI Act. Article 4 of the AI Act does not require any specific certificate. The course does not carry professional education credits.
When are the dates and how much does it cost?
The first S1 edition runs online in October 2026: two evenings of 2:30 h. Conference attendees with the code OKO pay PLN 450 net; enter the code in the enrolment form. We will confirm the exact session dates by email. Dates for the other courses are being set. Leave an enrolment and we will write once they are known.
Medical documentation and GDPR Article 9
of physicians' time in outpatient practice goes to documentation and desk work.
Special category data. GDPR Article 9 places health information among the most strictly protected data. Transcribing a consultation is processing of that data, with every obligation that follows.
Cloud tools process recordings on the vendor's servers, often with subprocessors. The institution has to cover each of them with a data processing agreement and assess them before any patient data reaches them. Responsibility for choosing the tool stays with the institution.
- 1/3 wizytyof a primary care consultation in Poland goes to documentation and admin (Supreme Audit Office, 2021)
- art. 9 RODOspecial category: health data
- 20 mln EURor 4% of worldwide turnover, if higher: the maximum fine for breaching GDPR Article 9 (Art. 83(5))
- 279 dniaverage time to identify and contain a healthcare data breach (IBM, Cost of a Data Breach 2025)
- AI Actobligations depend on the system's intended purpose and classification
One device. The whole consultation, in RAM.
No WiFi. No cloud. No exceptions.
The device has no WiFi module, no Bluetooth and no modem. The hardware is physically absent, so it cannot be switched on in settings. Local processing limits data transfers, but compliance also depends on how the device is used and on the safeguards in the institution.
Four steps. Nothing to learn.
The questions we get most.
Can I use my own microphone?
No. The device works only with its built-in microphone, calibrated for consulting-room acoustics and for telling two speakers apart. There is no input for an external microphone.
What is the USB port for?
Connecting to a computer. Over USB-C the device works with the MedScribe desktop app, where you see the transcript and the finished note. Power comes through a separate cable. The device sends nothing further.
Are there HIS integrations?
No, and that is deliberate. Every API integration adds an attack surface. The note goes to the system clipboard, and you paste it into mMedica, KAMSOFT, eDokumentacja, gabinet.gov.pl or any other system.
Is data encrypted?
There is nothing to encrypt, because nothing leaves the device. Audio is processed in RAM and overwritten once the note is generated. The local disk never stores audio or transcripts.
Does the device work without internet?
Yes, it is the only way it works. MedScribe has no WiFi module, no Bluetooth and no modem, so it cannot connect to anything. The transcript and the note are produced in the memory of the device on your desk.
Can Confivox see the content of consultations?
No. We have no technical means to. Neither audio nor transcript leaves the device, and the device has no way to send them. We collect no content telemetry and have no remote access.
Are recordings stored anywhere?
No. Audio lives in RAM only and is overwritten once the note is generated. Powering the device off wipes everything. No file is created that could leak, be seized or be replayed years later.
How accurate is the transcription?
The model is calibrated for Polish medical vocabulary, abbreviations and drug names, and diarisation separates the physician from the patient. We do not publish an accuracy figure yet. We measure it during the pilot programme, in a real practice, and share the results with pilot participants.
Can I edit the note before pasting it into records?
Yes. The SOAP note appears on screen for review and editing, and you decide what goes into the record. The device never writes anything into your clinical system by itself.
How does this work under GDPR and the AI Act?
Patient data never leaves the practice, so you entrust it to no one: there is no processor, no data processing agreement and no transfer outside the EEA. That simplifies your paperwork but does not replace it. The risk assessment and the record of processing activities stay on your side, ideally with your DPO. Our training supports AI literacy; further steps depend on the systems you use and on your institution.
How much does MedScribe cost?
We quote individually, as the price depends on the number of seats and the scope of support. The device is pre-launch and we are recruiting for the pilot programme. Send us a message through the form and we will reply with terms.