ARFA wires together the tools your practice already runs. Phone, fax, intake, scheduling, billing, documents, reporting. None of them were built to talk to each other. We build the workflows that bridge them.
Most operational pain in an independent practice comes from work that happens between systems your existing software doesn't connect. ARFA builds the connective tissue. Start with one workflow. Add the rest as confidence grows.
Every vertical below leaks money in a slightly different way. ARFA was built to find the leak and plug it. Click a card with a link to see the specific automations.
Synthetic patient data, BAA-gated stack, end-to-end traceable. Same pattern extends to every category above.
Every call answered on the first ring. Booking, refill, billing, or urgent triaged in under 2 seconds. Last-name plus date-of-birth verification before any patient info is disclosed. Emergency cases route to 911 with an audit trail.
If a call drops, rings out, or comes in after-hours, the patient gets a text back within 2 minutes. The lead lands in the dashboard automatically. Nothing leaks even when something upstream breaks.
Every appointment, refill request, callback, or escalation flows into the right staff queue. Categorized, tagged with an action verb, audit-logged. Front desk only sees scheduling. Clinical only sees medical. Billing only sees insurance.
I built ARFA on a small, deliberate stack. Every tool was chosen for a specific job and a signed Business Associate Agreement. Plain English on what each one does:
A voice platform that picks up the phone, listens to the patient, and has the actual conversation in real time. Verifies last name and date of birth before disclosing any patient information. Tested across 31 adversarial scenarios (social engineering, profanity, controlled-substance probes).
A production language model that classifies what each call is actually about (booking, refill, billing, urgent) and decides where it goes next. Bedrock is Amazon's enterprise hosting layer that comes with a signed BAA covering Claude.
An open-source automation engine that wakes the moment a call ends and routes the data: book in the scheduler, log to the dashboard, send the SMS, page the on-call clinician, parse a fax, route a denial alert. Idempotency keys mean a retry can't double-book. Latency budget under 800 milliseconds end-to-end.
A self-hosted database where patient information lives temporarily. Runs on our own DigitalOcean server. Every read and write is timestamped, user-tagged, and queryable. SOC 2 evidence-ready from day one.
The carrier-grade backbone that handles the actual phone calls and sends every SMS. Covered by Twilio's signed Business Associate Agreement.
Custom-built. What your team actually opens in the morning. Role-filtered (front desk only sees scheduling, billing only sees insurance). Drag-to-update status. Audit trail per row.
Every tool above was selected because it offers a signed Business Associate Agreement for HIPAA-covered data. Tools without a BAA, like standard public LLM APIs and open-routing aggregators, are never on the path.
Synthetic data only until then. No real PHI in any demo. The matrix below is exhibit A in every proposal.
| Tool / layer | Status |
|---|---|
| Production language model | SIGNED |
| Voice gateway | SIGNED |
| SMS gateway | SIGNED |
| Intake / scheduling | SIGNED |
| Compute / database host | SIGNED AFTER CONTRACT |
| Public LLM APIs | NEVER USED |
| Open-routing aggregators | NEVER USED |
Two tools at the bottom (public LLM APIs and open-routing aggregators) are never used for any practice we work with. They don't have BAAs. They aren't on the path.
I built ARFA after watching independent practices try to solve operational problems with software designed for hospital networks. The mismatch is everywhere. Practice management systems that don't know what the phone did. EHRs that can't see after-hours calls. Dashboards that show everything to everyone instead of routing what matters to who needs it. Fax workflows that still rely on someone walking a piece of paper across the room.
Most AI-for-healthcare pitches today are vibe-coded prototypes pretending to be production systems. ARFA is the opposite. Narrow scope, signed BAAs before the first real call, idempotent writes, dead-letter queues for failed parses, audit logs you can query. The dashboard your staff opens Monday morning is just the visible layer. Underneath is the discipline.
One operator, deliberate scope, no rip-and-replace. Pilots take 14 days. The math you'll get back when you reply to this page is your own. Your call volume, your patient lifetime value, your specific workflow leaks. No deck, no slides.
Your weekly call volume plus a rough estimate of how many go unanswered. You'll get back a leak estimate and the three workflows worth piloting first. No deck. 15 minutes if the math looks worth talking through.