AI systems for healthcare teams that need regulatory awareness from day one.
From health-data workflows to medical-AI validation, I help healthcare and medical-AI teams build AI that is useful, traceable, and ready for the GDPR, EU AI Act, MDR, and IVDR realities they operate in. Awareness and readiness are designed into the work from the start, not bolted on at the end.
What makes healthcare AI hard
The technology is rarely the blocker. These six pressures are what stall most healthcare AI work.
Health-data sensitivity
Patient and clinical data is among the most sensitive there is. Every design decision has to account for special-category data, minimization, residency, and the option to keep processing on-premise.
Unclear validation requirements
Teams often cannot tell how much evaluation, testing, and evidence a given use case actually needs. Too little and it fails review; too much and it never ships.
EU AI Act, MDR, and IVDR uncertainty
It is hard to know early whether a system is high-risk, whether it edges toward a medical-device classification, and what technical documentation that implies. Discovering this late means costly rework.
Integration with clinical and administrative workflows
AI only helps if it fits how clinicians and staff already work. Systems that ignore existing tools, records, and processes get quietly abandoned.
Procurement and information governance review
In a hospital the longest step is rarely the build. Information governance review, a data protection impact assessment, security questionnaires and a processor agreement all sit between a decision and a deployment, and they run in sequence unless somebody starts them early.
The pilot that cannot leave the pilot
A model validated on one site, one scanner or one clinic's coding habits often degrades on the second. Planning for that in the evaluation design is what separates a system that scales from one that stays a demonstration.
Where AI earns its place in healthcare
High-value, lower-risk use cases where careful AI delivers real benefit without overreaching.
Document intelligence
Extract and structure data from referral letters, lab reports, and discharge summaries, cutting manual entry while keeping data handling tight.
Triage and admin automation
Support routing, prioritization, and routine administrative work so staff time goes to patients rather than paperwork, with humans firmly in the loop.
Medical-image workflow support
Assist imaging and pathology workflows with measurement, sorting, and pre-processing that slot into existing PACS and DICOM pipelines.
Model evaluation and validation support
Build the evaluation harnesses, test sets, and metrics that let you trust a model's behaviour and produce the evidence reviewers expect.
Internal knowledge systems and RAG
Retrieval-augmented assistants over protected internal knowledge, designed with access control and data minimization so sensitive content stays contained.
Not sure which fits?
A Regulated AI Readiness Audit ranks these by value, feasibility, and regulatory exposure, so you start with the right one.
See the audit →A compliance-aware way to build
Healthcare AI works best when regulatory realities shape the design from the first sketch. The goal is not a pile of paperwork at the end, but systems that are easier to trust, review, and maintain.
This is technical and engineering support to help your team reach readiness. It is awareness-driven and aimed at building the right evidence and documentation, not legal advice or a substitute for your regulatory and legal counsel.
-
✓
Data minimization
Collect and process only what each use case truly needs, with residency and on-premise options where the data demands it.
-
✓
Traceability
Clear lineage from data to model to output, so decisions can be explained and reproduced when it matters.
-
✓
Evaluation and monitoring
Defined metrics, test sets, and ongoing monitoring to catch drift and regressions before they reach patients or staff.
-
✓
Technical documentation support
Help producing the technical documentation and evidence that EU AI Act and MDR/IVDR processes expect, ready for your reviewers and counsel.
-
✓
Human oversight
Systems designed to keep a qualified human in control of clinically meaningful decisions, with the AI in a supporting role.
Built by someone who has done the science and shipped the systems
I am a PhD computer scientist and former CTO, EU-based in Munich. My research was in medical imaging and histopathology, and I have built and run production AI systems, not just demos. That combination is the reason healthcare AI work here starts grounded in both the clinical science and the engineering reality.
Computer science
Technical leadership
Medical-imaging research
EU-based
Who we usually work with
Healthcare AI decisions are rarely made by one person. These are the four roles at the table, and the question each of them arrives with.
CTO and heads of engineering
“Can we build this with the team and the data we have, and what does it commit us to afterwards?” Usually the person who has to live with the architecture.
Data protection officers
“What leaves our environment, under what basis, and can we evidence it later?” Involving them after the architecture is chosen is the most common source of rework.
Clinical and medical directors
“Is it safe, does it hold across our patient mix, and what happens when it is wrong?” Accuracy in aggregate does not answer this; performance per subgroup does.
Product leads at medical software companies
“Does this make us a provider of a high-risk system, and what does that add to the roadmap?” The answer shapes the release plan more than the feature does.
Which engagement fits which project
Start with an audit when
The use case is named but not scoped, a vendor proposal needs evaluating, or nobody can yet say whether the system is high-risk or a medical device. The output is a decision, in writing.
The AI Readiness Audit →Run a proof of value when
The use case is defined and the question is whether it performs on your data, your sites and your patient mix. Set the threshold and the stop condition before it starts.
Scoped engagements →Bring in a fractional lead when
Something is already in production and needs an owner: evaluation standards, the regulatory posture, vendor decisions and the next three systems.
Fractional AI Lead →Worth reading before you commit
Where to go next
Whether you are scoping a first project or scaling a capability, there is a clear next step.
Regulated AI Readiness Audit
Decide which healthcare AI use cases are worth building, with validation and traceability mapped from the start.
Explore the audit →EU AI Act Compliance
Understand your risk classification and the technical documentation a high-risk healthcare system implies.
Compliance support →Fractional AI Lead
Ongoing senior AI ownership as you build and scale healthcare AI responsibly.
See the role →Writing and insights
Articles on building AI that is useful, traceable, and ready for regulated environments.
Read the blog →When should a healthcare company run an AI readiness audit?
A healthcare organisation should run an AI readiness audit before committing budget to a clinical or operational AI system, at the point where a use case has been named but not yet scoped.
The trigger is usually one of four situations: a vendor proposal nobody internally can evaluate, a clinical team asking for a tool that has not been costed, an open question about whether a system is high-risk under the EU AI Act or a medical device under MDR or IVDR, or a pilot that produced encouraging results but no path to production.
Running the audit first matters most where the constraints are hardest, and healthcare has the hardest ones: special-category patient data under the GDPR, validation evidence that has to hold across patient subgroups rather than in aggregate, and human oversight that must be meaningful rather than nominal. Each of these is expensive to discover after a build, and sometimes unrecoverable.
Frequently asked questions
Find safe, high-value healthcare AI
Start with a Regulated AI Readiness Audit. GDPR, EU AI Act, and MDR/IVDR awareness built in from day one.