How I Built an AI Tool That Processed 15,000 Prescriptions in 48 Hours

When I first heard about the charity healthcare event, the challenge was blunt: over 15,000 handwritten prescriptions needed to become structured JSON in less than 48 hours.

The Architecture

I built OmmSai with Python, Claude vision, Google Drive integration, ThreadPoolExecutor concurrency, and a Tkinter operator GUI so non-engineer volunteers could run the pipeline on Windows laptops.

def process_prescription(image_bytes):
    # Convert image to optimized base64
    # Call Claude with vision
    # Validate structured JSON response
    # Save extraction result

The durable lesson was that the problem was not OCR in isolation. The real problem was document AI under deadline pressure: structured extraction, confidence thresholds, human review, resumability, and operator UX.