Project N°02 / Healthcare AI

Vitalis Triage

Autonomous triage assistant integrated into EMRs.

02 · HEALTHCARE AI
Client
European Hospital Group
Year
2025
Duration
26 weeks
Team
8 engineers + 2 clinicians
— STAR breakdown
S

Situation

A 42-hospital European group faced record ER wait times and inconsistent triage decisions across sites, with clinicians drowning in unstructured EMR notes.

T

Task

Ship a HIPAA-compliant, FHIR-native triage assistant that reads structured and unstructured records, augments (not replaces) clinical judgment, and is auditable end-to-end.

A

Action

  • 01Fine-tuned a clinical LLM on de-identified EMR corpora with clinician-in-the-loop labeling.
  • 02Built a FHIR-native ingestion layer with on-prem inference for data residency compliance.
  • 03Ran a 12-week formal audit against clinician baselines across three pilot hospitals.
  • 04Rolled out with per-site kill switches and continuous drift monitoring.
R

Result

  • 94% triage accuracy vs. 87% clinician baseline in blinded evaluation.
  • 38% reduction in average ER wait time across pilot sites.
  • Live in 42 hospitals within 9 months of contract signature.
  • SOC 2 Type II + HIPAA certified; zero PHI incidents post-launch.
— Stack
PyTorchFHIRAWS