AI-Powered Patient Intake System

Overview

TopHealth AI runs a network of 23 urgent care clinics and came to us with a problem that sounds simple on the surface — patients spending too long at the front desk — but had years of bad process baked into it. We built an AI-powered intake system that extracts insurance information from photos and pre-fills patient records, cutting intake time from 12 minutes to under 4.

The Challenge

Every patient visit started the same way: a clipboard, a pen, and a stack of forms they'd already filled out a dozen times before. Front desk staff spent close to 40% of their day on data entry. Insurance verification was manual, error-prone, and causing billing delays that averaged 14 days per claim.

The existing EHR setup had no self-service intake path. There was no escape hatch — just more paper.

Lina, TopHealth's founder, put it plainly: the front desk had become a bottleneck that was affecting both patient satisfaction and staff morale. She needed a solution that worked on any phone, required zero app downloads, and could slot into their Epic environment without a rip-and-replace.

Our Approach

We spent the first two weeks not writing code. We shadowed front desk staff at three clinic locations to understand the actual intake flow — not the documented version, but what really happened under pressure with a full waiting room.

What we found: the biggest time sink wasn't the forms themselves, it was insurance verification. Staff were manually reading cards, typing policy numbers, calling payers. The AI opportunity was obvious.

From there we moved fast:

1. Mobile-first PWA — No app store friction. Patients access a Progressive Web App via QR code in the waiting room. Any smartphone, any browser.
2. OCR + LLM pipeline — A computer vision pipeline extracts data from insurance card photos with 94% accuracy. An LLM validates, structures, and flags edge cases before anything touches the EHR.
3. Epic integration via HL7 FHIR — Patient records update in real time. No manual re-entry, no dual systems.
4. Smart fallback layer — When AI confidence drops below 85%, the record is flagged for front desk review. This happens for less than 6% of submissions.

The Solution

The final system is deceptively simple from the patient's perspective: scan the QR code, photograph your insurance card and ID, confirm the extracted data. Done. You're checked in before you see the doctor.

Behind that simplicity is a pipeline handling OCR, entity extraction, validation, FHIR mapping, and Epic write-back — all in under 30 seconds.

Results

• Intake time dropped from 12 minutes to 3.8 minutes — a 68% reduction
• 40% of front desk time that was going to data entry got redeployed to actual patient care
• Insurance verification errors dropped 71%
• $340K in annual savings across 23 locations
• NPS increased by 18 points in the first quarter post-launch

What Lina Said

"Our front desk staff went from data entry clerks to patient advocates. DevNexus didn't just build software — they changed how our clinics operate."
— Lina, Founder, TopHealth AI