Document automation
Inbound faxes turned into radiology orders
A fax arrives, gets read, gets parsed, and lands in the imaging system as a structured order. No retyping.
- Client
- Diagnostic imaging practice
- Period
- 2025-2026
- Stack
- Python, PostgreSQL, HL7 v2, OCR, Vision models, systemd
The problem
Referrals arrived as fax images. Staff retyped every patient, every physician, every procedure code into the imaging system by hand. Backlog grew faster than the team could clear it, and every retyped field was a chance to enter the wrong patient.
What I built
Pipeline from the fax provider API to OCR, then to a field parser with geometry taken from the document text layer, then to an HL7 order message built against the imaging system. Every extracted field carries a confidence score, and anything below the line goes to a human review queue instead of into the system.
Result
The standing backlog was cleared and the queue now stays flat. Reversible by design: the pipeline never overwrites an order it did not create, and every automated decision is logged with its inputs.