MediCore Records Digitisation
Digitised 2 million paper patient records with 99.2% accuracy using OCR and a structured QA process — cutting record retrieval from 45 minutes to 8 seconds.
Client background
HealthCore Ltd. operates a network of 12 outpatient clinics across Dhaka and Chittagong, serving approximately 800 patients per day. The group had been operating for 18 years, accumulating over 2 million paper records stored across physical filing cabinets in each clinic's records room.
The challenge
Doctors were spending an average of 45 minutes per patient just locating their paper record folder before a consultation. Files were misfiled, lost, or physically deteriorating. When a patient visited a different clinic in the network, there was no way to access their history. The medical director estimated the record-retrieval overhead was costing the group 400 doctor-hours per week — time that could be spent on patients.
Discovery & planning
Initial Scoping Call
60-minute call with the Medical Director and Head of Operations. We assessed the volume (2M records), format diversity (handwritten, typed, mixed), and language (Bengali + English). Agreed on a phased approach.
Compliance & NDA
Reviewed Bangladesh Health Data Protection guidelines. Signed a mutual NDA covering patient data. Established a secure document transfer protocol — no records left clinic premises.
Sample Batch Test
Processed a sample of 5,000 records across 4 format types. Ran OCR + manual QA and reported accuracy by format. Handwritten records required higher manual QA ratio than typed ones.
Workflow Design Sign-Off
Presented a 5-stage quality control workflow. Client approved. Agreed on 99%+ accuracy SLA and a 14-week delivery timeline with weekly progress reports.
Technical solution
We deployed a 5-stage pipeline: (1) document scanning using clinic-owned scanners with a custom batch-scanning guide we produced, (2) OCR extraction using Tesseract with pre-processing for handwritten text, (3) structured data normalisation into a PostgreSQL schema with 47 standardised fields, (4) automated rule-based QA flagging for anomalies, and (5) a human QA team reviewing all flagged records. Processed records were uploaded to a secure AWS S3 bucket and made queryable via a simple search API delivered to HealthCore's IT team.
Tech stack
Timeline
Scanner guide, OCR pipeline setup, schema design, pilot batch (5,000 records), accuracy baseline
750,000 records — typed documents, highest OCR accuracy, QA team at 30% review ratio
1,000,000 handwritten records — elevated QA to 60% manual review, 4 additional QA operators onboarded
Remaining 250,000 mixed records, final accuracy audit, search API delivery, documentation handover
Results & outcomes
Finding a patient's full 10-year history now takes 8 seconds instead of 45 minutes. The accuracy is remarkable given how varied the handwriting quality was. This has genuinely changed how we care for patients.
Book a free call and we'll scope your project in 30 minutes.
Book a free call