EU AI Act · MDR · IEC 62304 · ISO 14971

Independent AI/ML Model Assessment
for Medical Technology

When your AI has to pass – not just perform.

Risk-based evaluation of AI/ML models with focus on data quality, validation, explainability and regulatory compliance – from development to approval.

Request Assessment View Services

The Problem

Many AI models look ready to ship — yet reveal gaps under regulatory scrutiny.

Hidden Data Issues

Unclear representativeness, labeling inconsistencies and undetected bias compromise model reliability.

Weak Validation

Insufficient test-train separation, missing external validation and overfitting risks remain undetected.

Lifecycle Gaps

Missing version control, unclear change management and no robustness strategy create long-term risk.

Left undetected, these gaps cause approval delays, costly rework and avoidable regulatory risk.

Who It's For

I work with organizations deploying AI in regulated medical environments – who need external expertise for technical assurance.

MedTech Scale-ups

Your data scientists build strong models – but the regulatory know-how for AI is missing. It works; is it also validated and documented to MDR standards?

SaMD startups, digital health, AI diagnostics

Established MedTech

You've mastered MDR – but AI models raise new questions. How do you validate a model that changes with every update? How do you document training data per IEC 62304?

Endoscopy, imaging, monitoring, implants

Investors & VCs

Your investment committee asks: "Is the AI market-ready AND approvable?" I deliver the technical due diligence for sound decisions.

Digital health VCs, life sciences funds

Services

Structured, independent assessments – the external perspective internal teams rarely have on their own model.

Quick Risk Check

Rapid assessment of your AI/ML model covering top risks, critical gaps and immediate recommendations.

  • High-level risk identification
  • Top 5 findings with traffic-light rating
  • Regulatory classification
  • Concrete next steps

Results in 5–10 business days

Full Technical Assessment

Comprehensive evaluation across all critical domains with detailed scoring and structured findings.

  • 8-domain scoring matrix
  • Detailed findings report
  • FMEA & risk classification
  • Explainability analysis (XAI)
  • Actionable recommendations

Scope by arrangement

Technical Due Diligence

Independent technical assessment for investors: Is the AI model technically sound, regulatory-viable and scalable?

  • Model & architecture review
  • Data quality & bias assessment
  • Regulatory readiness check
  • Structured investment report

3–5 days, confidential

Launch phase 2026

For the first pilot projects, I work closely and collaboratively with selected clients – at reduced rates. You receive a particularly intensive, hands-on engagement; I build referenceable case studies. A fair exchange for both sides.

Assessment Framework

Systematic evaluation across 8 domains – based on recognized standards and regulatory requirements.

01

Data & Governance

Quality, provenance, representativeness, bias analysis

02

Development & Training

Architecture, training design, reproducibility

03

Validation & Testing

Test strategy, subgroup analysis, metrics

04

Robustness & Risk

Generalization, failure modes, FMEA, stress tests

05

Explainability (XAI)

Interpretability, Grad-CAM, SHAP, clinical plausibility

06

Lifecycle & Updates

Versioning, change management, drift detection

07

Documentation

Technical file, model cards, traceability

08

Regulatory Compliance

MDR mapping, EU AI Act, risk classification

Referenced Standards & Frameworks

GMLP IEC 62304 ISO 14971 ISO 13485 EU AI Act MDR 2017/745 CLAIM Guidelines

How We Work Together

A clearly structured process – transparent, confidential and risk-free in the first step.

1

First Call

No obligation, free of charge. We clarify your needs, the context of your model and whether I'm the right partner for you.

2

Scoping & Proposal

After signing a non-disclosure agreement (NDA), we define scope, objectives and timeline – you receive a transparent fixed-price or time-and-materials proposal.

3

Assessment

Structured, independent analysis of your AI/ML model along the assessment framework – with interim alignments to keep everything grounded in your reality.

4

Report & Recommendations

You receive a clear findings report with prioritized risks and actionable next steps – including a results presentation on request.

About

Dr.-Ing. Ralf Lutchen

Dr.-Ing. Ralf Lutchen

PhD Data Scientist · Independent AI/ML Assessment

Doctoral-level Data Scientist (Dr.-Ing.) with over 15 years of experience in safety-critical technology and industrial software. My path bridges mechatronics, machine learning, IoT architecture and regulated product development – from commissioning real machines to production-grade AI systems.

As a Data Scientist, I build AI solutions at production level – meeting the demands on validation, reproducibility and robustness that an industrial environment imposes. My doctorate (graded 1.0, highest distinction) combined artificial intelligence, software architecture, IoT and cyber security: I developed a cloud-based AI method that reads vehicle networks and controls safety-relevant interventions.

This domain's requirements for functional safety, traceability and validation are closely related to those of medical technology. That is exactly the bridge I bring to RaMed: I transfer proven AI-validation methods from safety-critical industries to the regulatory requirements of medical technology – MDR, IEC 62304, ISO 14971 and the EU AI Act.

My approach follows established frameworks: CRISP-DM for data science workflows, arc42 for architecture documentation of larger systems. Extensive DevOps experience enables me to assess not just the model, but also data architecture, solution architecture and the full MLOps pipeline.

"I see what internal teams typically miss – the bridge between ML development and regulated reality."

Dr.-Ing. Data Science 15+ Years Safety-Critical Tech Safety-Critical AI Mechatronics & Software Explainable AI CRISP-DM arc42 DevOps & MLOps
View LinkedIn Profile

Frequently Asked Questions

What clients typically want to know before working together.

How is the confidentiality of my data and models ensured?

Before any substantive collaboration, we sign a non-disclosure agreement (NDA). Your data, models and documents are used solely for the agreed assessment, never shared with third parties, and deleted or returned after project completion on request.

Do you issue a certification or approval?

No. I provide an independent technical assessment and actionable recommendations. This does not replace review by notified bodies, authorities or certifying institutions – but my work prepares you specifically for it.

How long does an assessment take?

A Quick Risk Check delivers initial results in around 5–10 business days. A full Technical Assessment depends on scope and complexity and is defined together during scoping.

What does an assessment cost?

Pricing depends on scope and depth and is agreed transparently in advance, as a fixed price or based on effort. During the current 2026 launch phase, I work with selected pilot clients at reduced rates.

What documents do you need from me?

Typically information on data and data provenance, training and validation strategy, model architecture and existing documentation. In the first call we clarify what is genuinely useful – missing pieces are often part of the findings themselves.

Do you also work remotely / internationally?

Yes. Collaboration is predominantly remote; documents and alignment happen digitally and securely. The assessment is available in German and English.

Your AI model works
but will it pass an audit?

In a no-obligation first call, we gain clarity on your critical risks – in good time, before approval does.

Get in Touch