Case study Healthcare · New York · Custom AI tool

A DEXA scan used to end in a PDF.
Now it ends in a plan.

Prefusion Health runs three body-composition clinics across New York. We built them an iPad check-in, an AI reader that pulls every number off the scan, and a recommendation engine that speaks in the owner's own clinical logic. Live in all three locations.

Prefusion Health DexaFit NY
Bryan Jonas · Owner, Prefusion Health & DexaFit NY
Prefusion Health clinic, results review
Prefusion Copilot iPad check-in kiosk
Read per scan
14 metrics
Live in
3 clinics
14
metrics read off every scan
11/14
AI calls matched the clinician
35
check-ins, first weekend live
0
paper forms at the front desk
Manan Mehta briefing the AI Crew team in the Mumbai office
Built by real people
This one was scoped on a weekly call with Bryan, built by our crew in Mumbai, and working end to end on a real scan thirteen days later.
The AI Crew team at work
The brief, in Bryan's words

"Build a decision tool that takes the intake answers and the DEXA report and recommends the right protocol."

Weekly call, 19 June 2026
Paper intake form at the clinic
Before: a clipboard, a waiver on a separate system, and a scan report nobody could sell from.
01 · The situation

Great scan. Great data. No next step.

DexaFit NY does the testing, Prefusion Health does the treating. The scan itself was never the problem. Everything around it was.

description
Intake on paper

Demographics, safety screening and the waiver lived in three places. Every visit, every location.

picture_as_pdf
A scan that is only an image

The DEXA report exports as pictures, no text layer. Fourteen numbers a human had to read by eye.

groups
Technicians, not sales

The person handing over the results is a scan tech. Turning numbers into "here is your programme" was a coin toss.

02 · What we built

One iPad at the front desk. Three tools behind it.

Patient check-in kiosk
Tool 01
Patient check-in kiosk

New or returning, the patient taps through demographics, the safety screen and the real Prefusion waiver, signed in DocuSign. The physician order is generated the moment they finish. No clipboard.

AI scan reader output
Tool 02
AI scan reader

The tech saves the scan to a shared folder. The kiosk sees it, a vision model reads all fourteen metrics off the pages, and flags what matters: focus, watch, healthy.

Package recommendation toggle
Tool 03
Recommendation engine

Bryan's own clinical rules pick the package and explain why in the patient's numbers. The tech can flip between packages live, and the patient walks out with an A4 report.

account_treeHow the Copilot works

From walk-in to walk-out, on one iPad.

What the patient does, what the AI does and what the staff do, in the order it happens at the front desk.

Comes in
tablet_macPatient at the iPadNew, returning or staff mode
drawWaiver, signedIn DocuSign, before the scan
DEXA reportSaved to a shared Drive folder
monitor_heart

Prefusion Copilot

Reads the scan and applies the owner's own clinical rules.

  • check_circleA vision model reads 14 metrics off every page
  • check_circleFive clinical scenarios pick the package
  • check_circleThe model writes the why, never the pick
Goes out
picture_as_pdfA4 take-home reportThe plan, in the patient's numbers
descriptionPhysician orderGenerated as check-in finishes
Staff portal recordEvery check-in logged
Patient
touch_app01Walk in, tap the iPad

New patient, returning patient or staff mode. Returning patients pull their file by email.

Patient
draw02Screen and sign

Demographics, the safety questions and the waiver, signed on the spot.

Technician
accessibility_new03Run the scan

The DEXA runs exactly as before. Nothing changes in the scan room.

Kiosk
04Report lands in Drive

The tech saves the PDF to a shared folder. The kiosk picks it up by itself.

AI
visibility05Read, then decide

A vision model reads every metric. The owner's rules pick the package, server side.

Tech and patient
assignment_turned_in06Walk out with a plan

The tech flips through packages live. The patient leaves with an A4 report.

Built withtablet_maciPad kioskvisibilityVision modelruleRules engineGoogle DrivedrawDocuSignSupabasepicture_as_pdfA4 report
03 · How it works

From walk-in to walk-out, eight steps.

Everything the staff already did, in the same order. The AI sits inside the flow, it does not replace it.

1Walk in, tap the iPad

New patient, returning patient, or staff mode. Returning patients pull their file by email.

2Screen and sign

Demographics, the safety questions, the questionnaire, then the waiver signed on the spot.

3Physician order, done

The visit logs to the portal and the signed physician order PDF is generated automatically.

4The scan

The technician runs the DEXA exactly as before. Nothing changes in the scan room.

5Report lands in Drive

The tech saves the PDF to a shared folder. The kiosk polls it and picks up the new scan by itself.

AIVision reads the scan

Every page goes to a vision model. Body fat, lean mass, visceral fat, A/G ratio, ALMI, T-score and more come back as numbers.

AIRules decide the package

Bryan's scenarios run server-side on the numbers. The model writes the explanation, it never picks the package.

8Walk out with a plan

The tech flips through packages live, the patient leaves with an A4 report, the portal keeps the record.

04 · The intelligence

The numbers decide. Not the model.

Bryan already knew what a high visceral fat plus a high A/G ratio meant. He had just never had it written down as a rule. So we sat with him and turned his clinical instinct into five scenarios the tool checks on every scan, before any AI writes a word.

Visceral fat high · A/G ratio high
Central fat with inflammation markers
Prefusion Protocol + labs
Visceral fat high · A/G ratio normal
Storage without the inflammation signal
Optimization + Zone 2
Low lean mass · low visceral · low A/G
The muscle question, not the fat question
Protocol, bloodwork first
RER above 0.8
Burning sugar, not fat
Optimization + RMR
Clean scan
Nothing to fix, something to keep
Flex 4-Pack

Each package still gets its own personalised pitch in the patient's numbers, so the tech can pivot live when someone says "I do not want labs".

Recommended package view
Alternative package view
What Bryan asked for next

"Each package should have a personalization element to it. That way we can kind of flip through it."

Bryan Jonas, owner · Loom feedback, July 2026 · shipped the same week
05 · What changed

Same clinic. Same staff. Different conversation.

Before
  • closePaper intake, waiver on another system, order form typed by hand
  • closeScan report read by eye, results explained from memory
  • closeUpsell depended on who was on shift
  • closeNo record of who was recommended what
After
  • checkOne iPad, one flow, physician order generated automatically
  • checkFourteen metrics read in seconds, flagged focus / watch / healthy
  • checkEvery tech pitches the same package for the same numbers
  • checkTake-home A4 report, and every check-in logged in the staff portal
tablet_maciPad kiosk visibilityVision model ruleRules engine Google Drive sync drawDocuSign waiver Supabase picture_as_pdfA4 report
Timeline
  • 19 June · The brief
    One line on the weekly call. Intake plus scan plus journey, recommend the protocol.
  • 2 July · Working system
    Kiosk, Drive auto-pull, vision reader and A4 report tested end to end on a real scan.
  • 16 July · Handover
    SOP kit and QR codes for all three locations. AI graded against the clinician's own calls: 11 of 14.
  • 20 July · Live
    35 check-ins across Bellmore, Manhattan and Stony Brook in the first weekend.
  • Since then
    Package toggle, five clinical scenarios, personalised pitches. Now extending into the TRT journey.
Technician at the DEXA console
Prefusion Health clinic
Staff post-scan mode
The AI Crew team in Mumbai
Case studies · built by the crew

More case studies. All live.

All case studies

Have a report your team cannot sell from?

Clinics, labs, advisors, anyone whose value is buried in a PDF. We build the tool that turns it into the next step.

Book a callarrow_forward