MediScan AI

The problem

A chain of diagnostic clinics was manually re-keying patient intake forms and lab requisitions into their HIS - 6 minutes per form, 200+ forms a day, with a 4% error rate.

What we built

Jora built a computer-vision pipeline (Python + FastAPI) that ingests scanned forms, extracts structured fields with a fine-tuned layout model, validates against the HIS schema, and routes uncertain reads to a human-in-the-loop queue.

The outcome

Average processing time dropped from 6 minutes to under 30 seconds per form, the error rate fell below 1%, and the front-desk team reclaimed ~18 hours/week.