Independent AI/ML medical model audit

Regulatory-facing evidence support for medical AI systems

I help MedTech, digital health, pharma, clinical AI, and biomedical research teams evaluate whether their AI/ML models are scientifically valid, reproducible, explainable, bias-aware, and ready for clinical, investor, production, or regulatory-facing decisions.

From SaaS to on-device AI

Technical audit before claims become risks

Many medical AI failures are not architecture failures. They are evidence failures: weak validation design, hidden leakage, confounding, poor calibration, unstable features, subgroup blind spots, irreproducible notebooks, or deployment assumptions that do not match the intended use.

Who this is for

MedTech and SaMD startups

Independent technical review before investor due diligence, clinical pilots, certification preparation, or production deployment.

Angel investors, VC funds, and family offices

Independent technical due diligence on medical AI, SaMD, digital health, biomarker, LLM/CDSS, and on-device AI startups before investment decisions.

Digital health and SaaS AI teams

Review of cloud-based diagnostic, prognostic, triage, clinical decision-support, and monitoring systems.

On-device AI developers

Assessment of compact, edge, ultrasound, wearable, mobile, and embedded models where compute, data, and deployment constraints matter.

Pharma, CROs, and clinical research groups

Audit of biomarker discovery, omics pipelines, patient stratification, clinical prediction, and multimodal biomedical AI claims.

For investors and technical due diligence teams

Independent review of medical AI startup claims

I support angel investors, venture capital funds, family offices, corporate innovation teams, and strategic partners with independent technical due diligence on medical AI, SaMD, digital health, biomarker, clinical decision-support, and on-device AI companies.

The review focuses on whether the startup's AI claims are scientifically credible, reproducible, clinically meaningful, bias-aware, and aligned with the claimed intended use. This may include assessment of validation design, data leakage, external validation, subgroup performance, calibration, explainability, deployment assumptions, regulatory-facing evidence gaps, and whether reported performance is likely to generalize beyond the development dataset.

Investor Due DiligenceMedTech AISaMD StartupsDigital HealthBiomarker CompaniesClinical AILLM/CDSSOn-Device AI

Request investor due diligence review

What I audit

  • Intended use, target population, clinical workflow, and claim consistency.
  • Data provenance, cohort construction, inclusion/exclusion criteria, label quality, and missingness.
  • Train/validation/test split design, site-level separation, temporal leakage, patient leakage, and preprocessing leakage.
  • Confounding, batch effects, spectrum bias, subgroup performance, and fairness-relevant failure modes.
  • Metrics, thresholds, calibration, confidence intervals, decision curves, sensitivity/specificity trade-offs, and uncertainty.
  • External validation strategy, distribution shift, robustness, monitoring assumptions, and model lifecycle risks.
  • Explainability, feature stability, biological/clinical plausibility, and biomarker interpretability.
  • Reproducibility of notebooks, code, random seeds, environment, data transformations, and reporting.

Structured engagement options

Audit packages

Audit Snapshot

1-2 weeks

Fast independent review of model validity, leakage risks, metrics, documentation gaps, and next-step priorities.

  • Short technical report
  • Risk register
  • Priority fixes

Full ML Pipeline Forensics

4-8 weeks

Deep review of data, code, splits, features, training, validation, XAI, bias/confounding, and reproducibility.

  • Detailed audit report
  • Reproducibility review
  • Evidence-quality assessment

Regulatory-Facing Evidence Review

2-4 weeks

Technical gap analysis of evidence, intended-use alignment, clinical validation, model documentation, and reporting claims.

  • Evidence gap map
  • Validation roadmap
  • Documentation recommendations

SaaS / On-Device AI Readiness Review

3-6 weeks

Review of deployment, monitoring, drift, calibration, model-size constraints, low-resource inference, and operational risks.

  • Deployment-risk review
  • Monitoring assumptions
  • Lifecycle recommendations

Investor / VC Technical Due Diligence Review

1-3 weeks

Independent assessment of a medical AI company's technical claims, evidence quality, validation maturity, deployment assumptions, and scientific credibility.

  • Technical claim review
  • Evidence-risk summary
  • Investment diligence questions

Relevant medical AI domains

  • Diagnostic and prognostic medical AI.
  • SaMD, SaaS clinical AI, and clinical decision-support systems.
  • LLM/RAG-based healthcare tools and adverse drug reaction workflows.
  • Biomarker discovery, omics, MLE-NITs, and patient stratification.
  • Medical imaging, ultrasound, compact vision transformers, and on-device AI.
  • Survival analysis, risk prediction, calibration, and subgroup validation.

Typical outputs

  • Independent model audit report.
  • Validation and metrics review with technical risk register.
  • Leakage, confounding, and reproducibility assessment.
  • Regulatory-facing evidence gap analysis.
  • Scientific-methodology review for manuscripts, investors, or technical due diligence.
  • Investor-facing technical due diligence memo on medical AI startup claims and risks.
  • Actionable recommendations for model repair, validation, and documentation.

Important scope note

Scientific and technical support, not legal regulatory certification

I provide independent scientific, statistical, and machine-learning evidence support for regulatory-facing, clinical, investor, and production-readiness decisions. This work does not replace notified bodies, legal counsel, quality-management specialists, or regulatory affairs consultants.

Start with a short scoping message

Request an AI/ML medical model audit discussion

Send a short description of your model, data modality, intended use, current validation stage, deployment context, and timeline. I will respond with the most appropriate audit format.

Direct contact

Prefer email?

For confidential or early-stage projects, send a short non-confidential summary first.

ath.angelakis@gmail.com