MDR · IVDR · EU AI Act
You have machine learning in your device — and a notified body that has never seen anything quite like it. I help manufacturers of medical and IVD devices get AI-based products through conformity assessment, and keep them compliant as the models evolve.
Prof. Dr. Oliver Haase · Konstanz, Germany · working in English and German
Why this is hard
The frameworks were written for deterministic software. Three gaps come up in almost every project.
MDR has no equivalent of the FDA's Predetermined Change Control Plan. Every retrained model risks being treated as a significant change — and another external assessment.
The EU AI Act applies on top of MDR and IVDR, not instead of them. Risk management, data governance and post-market monitoring have to satisfy both without duplicating your QMS.
Clinical evaluation, performance validation and risk controls for a neural network look different from classical software. The burden is on you to explain them — in documentation that holds up in an audit.
Services
From a half-day briefing for your team to sustained work on a technical file.
Developing AI-based medical and IVD devices in compliance with legal requirements — a practical seminar for development, regulatory and quality teams. Delivered online or on site, in English or German.
From €2,500 — half day, up to 50 participants
A focused review of where your AI-based device stands against MDR/IVDR and the AI Act. You get a written report with prioritised findings and a realistic route to closing them — a good first step before committing to a larger engagement.
Fixed price — scoped after a first call
Hands-on support inside your project: change control for continuously learning models, technical documentation, risk management for ML, and the arguments you will need in front of your notified body.
Day rate or fixed scope — on request
An independent, written assessment of a specific question: is this feature a medical device, does this change require reassessment, is this validation evidence sufficient? Useful for internal decisions and for discussions with authorities and auditors.
Fixed price — scoped after a first call
References
Trusted by industry leaders
Who you work with
Validate ML is deliberately small. You work with me directly — not with a team you meet once at the kick-off.
I am a professor of computer science and a regulatory consultant, which is an unusual combination and the point of the whole thing: I can read your training pipeline and your technical file in the same afternoon. Most regulatory consultants cannot assess whether a validation strategy is scientifically sound; most ML engineers cannot tell you what a notified body will accept. The interesting problems live exactly in that gap.
Tell me what you are building and where it is stuck. A 30-minute call usually makes it clear whether the problem is small, large, or not really a problem at all.