The challenge
A multi-site hospital network received thousands of referral letters, lab reports, and discharge summaries each month — as PDFs, faxes, and scans. Clinical staff manually read, classified, and re-keyed them into the EMR, creating multi-day backlogs that delayed care decisions. Public AI services were ruled out: patient data could not leave the network’s infrastructure.
Our approach
We deployed a private document-intelligence pipeline inside the hospital’s own VPC: open-weight LLMs for clinical document understanding, OCR for scanned input, and direct integration with the existing EMR. Every extraction carried a confidence score; low-confidence items routed to a human review queue with the original document side by side.
What we built
- Private LLM deployment (VPC, no external API calls) for clinical document understanding
- OCR + extraction pipeline for referrals, lab reports, and discharge summaries
- Direct EMR integration with audit logging on every write
- Human-in-the-loop review queue for low-confidence extractions
- HIPAA-aligned access control and full interaction audit trail
Results
- Document turnaround dropped from 2–3 days to under 15 minutes for the standard path.
- Clinical staff reclaimed an estimated 60+ hours per week across sites.
- Zero patient data left the hospital’s security perimeter — verified by network audit.
- Extraction accuracy exceeded the manual re-keying baseline within 8 weeks.
"The deciding factor was that nothing leaves our infrastructure. We got modern AI without a single compromise on patient privacy."
Related services
Illustrative engagement scenario representative of Primis delivery work, anonymized for client confidentiality.