Platform
Four layers, one record
Medici is the platform. Orbis is the regulatory intelligence engine inside it. Together they put clinical development, safety, quality and regulatory affairs on one data model, one identity for every subject, site, product and document, and one audit trail.
How the layers fit
1. The data foundation
CDISC-native by design rather than by conversion. CDASH, SDTM, ADaM and SEND are enforced at the point of capture, not at submission, with controlled terminology (MedDRA, WHODrug, SNOMED, LOINC, ICD) and an immutable audit trail on every transaction.
Exchange runs through HL7 FHIR R4, OMOP for real-world data, and open REST and GraphQL APIs. Data lineage is tracked from source to submission artefact.
2. The process engine
Every procedure runs as a defined, monitored flow with human checkpoints, timers, escalation and a complete record of who did what and on what evidence. Sixty processes and 10,239 stages are configured today9.
3. The intelligence layer
AI operates inside those defined processes, so it has context, boundaries and an auditable record. It never approves, submits or releases on its own authority. What the AI does and does not do
4. The reporting layer
Because everything shares one model, a cross-programme view is a query rather than a data warehouse project.
Orbis, the regulatory engine
Orbis holds what each regulator requires and checks live records against it, so submissions are assembled from the record rather than rebuilt by hand.
- Monitoring of regulatory change with automated impact assessment against live submissions, protocols and SOPs
- Parallel submission planning across jurisdictions, with dependencies mapped
- Accelerated pathway eligibility: Fast Track, Breakthrough, PRIME, SAKIGAKE and ILAP
- A regulatory commitment register: every commitment to every health authority, with due dates and owners
- Dossier gap analysis across simultaneous submissions
- Reportability and deadline determination for safety events, per jurisdiction
AI with boundaries
Models classify, draft, extract and score inside defined processes. A qualified person decides.
What the AI does
- Drafts MedDRA and WHODrug coding for medical review
- Classifies trial master file documents and flags completeness gaps
- Scores literature for clinical evaluation reports
- Assesses the impact of new regulatory guidance on live programmes
- Prioritises data queries and detects anomalies in clinical data
What it never does
- Approve a submission
- Release a payment
- Issue a certificate
- Act outside a recorded process
Every AI action is logged with its inputs, its output, the process stage it ran in and the person who accepted, amended or rejected it. Any AI-assisted change is reversible.
Built for the people who will inspect it
Inspection readiness stops being a project and becomes a property of the system.
Audit by default
Every record and every process action carries who, what, when, the previous value and the new one.
Validation and data integrity
Designed to the electronic records and signatures expectations of 21 CFR Part 11 and EU Annex 11, within quality frameworks including ICH E6(R3) GCP, ISO 13485, ISO 14971 and IEC 62304.
Access control
Role-based access enforced at field level, attribute-based rules for multi-sponsor environments, single sign-on through SAML 2.0 and OpenID Connect, and automated provisioning.
Data residency
The tenant can be hosted in the region a client's policy or a funder's condition requires, with transfer restrictions enforced by geography.
Separation between sponsors
A multi-tenant architecture with complete data isolation between sponsors, and governed cross-tenant workspaces only where both sides agree.