Convened by ITU · WHO · WIPO · GI-AI4H
Open Code Infrastructurefor trustworthy health AI
A unified platform for the Global Initiative on AI for Health — catalog, annotate, evaluate, and report on AI in clinical and public-health workflows, end to end.
Platform · 12-18 month roadmap
Four workstreams, one provenance chain
Every artefact — dataset, annotation, evaluation, report — points back to the dataset that fed it. So a regulator can trace any claim to its evidence, and a host can revoke a downstream artefact if consent is withdrawn upstream.
Phase A · Catalog
Discover datasets curated under the GI-AI4H Topic Groups. Croissant 1.1 + RAI + BIOCroissant manifests, with provenance and consent tracked end-to-end.
Phase B · Annotation
Coordinate multi-rater annotation with conflict resolution, regulator-grade audit trails, and burndown tooling.
Phase C · Evaluation
Evaluate health AI models against benchmarking tasks without publishing the reference labels: predictions scoring, sealed-container execution, and a reviewed registry of evaluation methods.
Phase D · Reporting
JSON-LD evaluation reports for regulators, plus a portal that lets supervisors trace every claim back to its evidence.
For dataset hosts
Publishing on the catalog takes two steps: create a dataset, attach a Croissant 1.1 manifest. We validate the manifest against Croissant + RAI + BIOCroissant before mirroring distributions, so what lands in the catalog is machine-readable and audit-ready.