ClearPath Health operates 12 private hospitals across the US. Their ER intake process was entirely paper-based: patients filled out forms at reception, a nurse manually reviewed them and decided urgency level. During peak hours, this created a bottleneck that left high-acuity patients waiting behind low-acuity ones.
The Challenge
Build a HIPAA-compliant mobile triage tool that patients could use before or on arrival. It needed to handle 300+ symptom combinations, support English and Spanish, and integrate with their existing Cerner EHR without a full API contract.
Our Approach
We fine-tuned GPT-4o on a labelled dataset of 14,000 anonymised triage cases provided by the client's medical team. The model classifies symptoms into 5 urgency tiers aligned with the Emergency Severity Index. A React Native app handles intake; a FastAPI backend processes classification and writes structured output to a staging layer that feeds Cerner via HL7.
Results
After a 6-week pilot across 3 hospitals and full rollout to all 12, average ER wait times dropped by 38 minutes. Triage accuracy reached 91% against the nurse gold standard — exceeding the 85% contractual threshold. The app now processes over 47,000 patient intakes monthly.
The AI routing alone has paid for the entire project three times over in staff hours. Our nurses focus on care, not paperwork.