What the client was up against.
Clinicians were losing two hours per shift to documentation and triage inefficiencies.
Previous AI initiatives had stalled in model-risk review, with no clear path to clinical endorsement.
What we built, and how.
BNS delivered a FHIR-native data foundation with clinical-grade lineage and a retrieval-grounded assistant bounded by clinical policy.
Evaluation harnesses were signed-off with a joint clinical and model-risk panel; every prompt, response and decision is logged.
A human-in-the-loop workflow keeps clinicians in control of every triage decision.
How the program ran.
- Phase 01
Safety design
Clinical policy, acceptance criteria and evaluation harness.
- Phase 02
Data platform
FHIR ingestion, lineage and governed retrieval.
- Phase 03
Pilot
Two-hospital pilot with blinded clinical evaluation.
- Phase 04
Rollout
42-hospital rollout with ongoing drift and safety monitoring.
“The platform passed our regulator audit on the first pass. That is unprecedented for a clinical AI rollout at our scale.”