Pioneer Health, an Australian general practice, receives a high volume of inbound clinical documents each day, much of it scanned paper such as referrals, specialist letters and reports. Each document had to be read and manually keyed into their Best Practice clinical system, a slow and error-prone task that pulled staff away from patient care. Infostatus delivered a serverless pipeline on AWS that automates this securely, with all processing kept within Australia.

The challenge

• Staff manually identified the patient, date of birth and referring doctor on every document and typed the details into Best Practice
• Inbound files ranged from clean digital PDFs to poor scans and handwriting, defeating simple OCR
• As health data, all processing had to remain within Australia with a complete audit trail
• The output had to drop straight into Best Practice with no change to the clinical workflow

What we did

A fully automated, event-driven pipeline turns an uploaded document into a structured record ready for Best Practice.

1. Secure upload. Clinics upload documents through a secure interface into Amazon S3.
2. Primary extraction. Amazon Textract performs OCR and field extraction, and high-confidence results are accepted automatically.
3. AI fallback. Low-confidence documents are routed to Amazon Bedrock with Claude for human-like understanding of complex or handwritten layouts, entirely in-region.
4. Delivery. The extracted data is formatted into standard HL7 messages, with the original document attached, and delivered straight into Best Practice.
5. Quality and monitoring. Every document is logged with per-field confidence scores, failed extractions are flagged for review, and administrators are alerted to issues in real time.

The pipeline is built with AWS Lambda and AWS Step Functions, with Amazon CloudWatch for monitoring, and provisioned end to end with Terraform.

Outcomes

• Document handling reduced from hours to minutes
• High extraction accuracy, with AI handling only the hardest documents
• Full Australian data residency, with no international data transfer
• A very low, serverless per-document cost that scales linearly with volume
• A complete, timestamped audit trail for every document
• A prototype that has since grown into a live product used by multiple practices

“Infostatus did a great job for us. We came to them needing an AWS prototype for an AI document handling system, and they turned it around quickly. Their practical knowledge and experience gave us clean, efficient code we could build directly from. That saved us months of development time. We never could have moved as fast without the work they put in. We’ve gone from a simple prototype to a fully running product, with dozens of customers and more coming in.”

Andrew Walker, Head of IT Support, Pioneer Health