Healthcare

Posture

Healthcare is the vertical where AI's promise and AI's regulatory surface are both at their largest. Clinical decision support, imaging interpretation, ambient documentation, patient-facing AI, operational optimisation, drug discovery: each is a workload AI is materially improving. Each is also a workload where the regulatory frame is deeper than any other vertical: medical-device regulation, data-protection regulation, professional-practice regulation, and AI-specific regulation overlap, and the overlap is the engineering surface.

BtSavvy's posture on healthcare is regulation-first, engineering-deep. Healthcare AI cannot be retrofitted into compliance: the regulatory frame demands compliance evidence the delivery system itself emits, and FORGE is built around precisely that posture.

Regulatory posture

EU AI Act: many medical-device AI systems are designated high-risk under Article 6 and Annex III, triggering Article 14 human-oversight requirements, conformity-assessment procedures, and post-market monitoring. Devices already CE-marked under MDR or IVDR receive integrated conformity-assessment treatment under the AI Act (counsel to confirm against the enacted text).

EU MDR (2017/745) / IVDR (2017/746): where AI software meets the medical-device definition, it is regulated as a medical device, with classification rules, conformity assessment, post-market surveillance, and unique device identification; such software cannot be placed on the EU market without MDR/IVDR conformity. Much healthcare AI (operational optimisation, many ambient-documentation and decision-support tools) falls outside the device definition, and getting that boundary right is the first engineering decision.

US FDA — Software as a Medical Device (SaMD): the FDA regulates AI/ML-enabled medical devices through the SaMD framework, with guidance on Predetermined Change Control Plans for adaptive systems and Good Machine Learning Practice.

HIPAA (US): applies to protected health information, with specific implications for AI systems that ingest, infer about, or output PHI.

UK GDPR / EU GDPR plus health-data special-category provisions: health data is special-category under Article 9, with stricter lawful-basis and safeguard requirements.

ISO 13485: medical-device quality-management-system standard. ISO/IEC 42001: AI Management System standard.

Which FORGE services carry the most weight in healthcare

Assurance. The load-bearing service: composing the conformance evidence the applicable regimes require, continuously: EU AI Act, MDR/IVDR, FDA SaMD, or HIPAA, per the workload's market and device classification.

Data Foundation. Healthcare data is heterogeneous, special-category, and frequently consent-bound: clinical, claims, imaging, and patient-reported data brought into agent-ready form under lineage and consent discipline.

Strategy & Architecture. Typically establishing the medical-device boundary, the clinical-validation pathway, and the deployment-and-monitoring posture.

Run. SRE and managed operations calibrated against clinical-impact failure modes, with hybrid automation and clinician-in-the-loop discipline.

How FORGE works in healthcare

Factory concentrates on accelerators calibrated against healthcare failure modes: clinical-decision-support calibration regression, imaging-model performance monitoring across patient demographics, ambient-documentation accuracy regression.

Observability includes performance-by-subgroup observability: model performance broken down by demographic, clinical-context, and care-setting axes, monitored continuously rather than measured once at validation.

Review includes regulator-facing and clinician-facing review. The signed gate log surfaces evidence directly to MDR/IVDR notified bodies, FDA reviewers, and clinical-governance committees.

Governance and Trust treats the medical-device boundary as policy-as-code. AI inside the boundary carries MDR/IVDR or FDA discipline; AI outside it carries HIPAA, GDPR, and EU AI Act discipline. The boundary is enforced architecturally.

Expert Intelligence combines frontier-model competence with healthcare-domain expertise: clinical-language fine-tuning, retrieval against clinical guidelines and evidence bases, calibration against subgroup-fairness criteria.

Use-case sketches

A clinical-decision-support deployment that emits EU AI Act conformance evidence continuously. A frontier-model-augmented decision-support system for a clinical pathway, with retrieval-quality and calibration-skew telemetry as live SRE signals, the EU AI Act Article 14 human-oversight controls enforced as a promotion gate via policy-as-code, and a traceability genome connecting each clinical recommendation to the underlying evidence.

An ambient-documentation deployment with HIPAA-compliant audit trail. An ambient clinical-documentation system with full traceability from speech to draft to clinician-edit, signed at the clinician-acceptance gate, and HIPAA-compliant audit logging.

An imaging-model deployment with subgroup-performance monitoring. A medical-imaging interpretation system with continuous performance monitoring by demographic and clinical-context subgroup, named-engineer approval at every model promotion, and a regulatory-submission package emitted by the delivery system itself.

Insights cross-links