Healthcare

AI agents for the work your EMR leaves behind

Prior auths sit in a fax queue. Referrals get retyped into three systems. Documentation waits until after the shift. We build agents that read the clinical data, act in the EMR, and hand the genuine edge cases to a human. HIPAA compliant, audit-logged, in production.

prior_auth.runHealthcare
  1. intake.readfax_2291.pdf → structured request
  2. policy.checkCMS-1500 against payer rules
  3. emr.update3 records written back
  4. queue.route1 edge case escalated to a human
Example run · 4 calls · 3 systems · 1 human decision

Trusted by teams building and running mission-critical platforms

Dunn
HCA
Healthstream
Novo
Accureg
Challenges

Where the hours go in healthcare

Nobody in a health system is short of process. They’re short of people to run it. Each of these is a queue somebody is clearing by hand today.

01

The fax machine is load-bearing

Prior auths, referrals, and orders arrive as PDFs and faxes. Someone reads them, retypes them into the EMR, and chases the payer. Highest volume, lowest judgment work in the building.

02

Every integration is a person

The EMR doesn’t talk to the payer portal, which doesn’t talk to scheduling. Staff hold it together across browser tabs, and the handoffs are exactly where the errors and denials come from.

03

Denials get worked, not prevented

By the time a claim comes back denied, the documentation gap is weeks old. Nobody checked eligibility and medical necessity before the service happened, because checking by hand doesn’t scale.

04

The pilot stalls at compliance

The demo worked. Then legal asked where PHI goes, who approved the output, and how you would prove either one in an audit. The project went back on the shelf.

Delivery

Design, build, and run — in your stack

Nothing here asks you to replace the EMR, retrain the floor, or wait out a platform program. One senior squad takes it from the first workshop to agents running in production.

Discover01

Count the touches

We sit with the people doing prior auth, intake, and documentation and count every touch. You get workflows ranked by volume, judgment required, and denial exposure — including the ones we tell you not to automate.

Design02

Scope, threshold, escalation path

Every agent gets a defined scope, a confidence threshold, and a named human to escalate to. Where PHI travels and what the BAA covers get settled at this stage, while they are still cheap to change.

Build03

Inside the EMR, not beside it

We write back through the interfaces you already own — HL7, FHIR, and the fax gateway. No rip and replace, no second system for staff to learn, no swivel-chair between the agent and the chart.

Manage04

We run it until your team owns it

Monitoring on every run, accuracy tracked over time, and an engineer on call. Handover happens when your team is ready to take it, not when the engagement ends.

Systems we work inside

Agents act in your systems of record. No parallel tool for your team to maintain.

  • Epic, Cerner, and Meditech
  • HL7 v2 and FHIR interfaces
  • Payer portals and eligibility APIs
  • Fax gateways and document intake
  • Azure Health Data Services
  • Scheduling and revenue cycle platforms
Use cases

What we ship in healthcare

None of these answer questions about the workflow. They run it: read the document, call the systems, write the result back, and escalate what a clinician should genuinely decide.

Prior authorization

Reads the inbound fax or portal request, checks it against payer policy, submits, and tracks status to a decision. Escalates only when the policy is genuinely ambiguous.

Fax gateway → payer API → EMR

Referral intake and routing

Turns an unstructured referral into a structured record, matches it to the right specialty and site, and books or queues it without a coordinator retyping anything.

Intake → EMR → scheduling

Eligibility and necessity checks before service

Scores reimbursement likelihood against the specific Medicare and Medicaid rules that govern the service, and flags the missing documentation while it can still be collected.

EMR → policy layer → eligibility API

Clinical documentation reconciliation

Compares the note, the order, and the claim, finds where they disagree, and writes the corrections back with the source of each change attached.

EMR → document store → claims

Denial root-cause and appeal drafting

Groups denials by actual cause instead of payer code, drafts the appeal with the supporting record cited, and reports the pattern worth fixing upstream.

Claims → EMR → appeal packet

Proof

We’ve done this already

Numbers from delivered engagements, not a marketing deck.

Booked through a scheduling platform we built on top of a legacy EMR

250K+

appointments a year

One health system, several hundred call center and front desk staff

100+

clinics live

Held across peak volumes, including open enrollment

99.9%

uptime

Read the scheduling case study
Start here

Bring us your worst queue

Prior auth, intake, the documentation backlog — whichever one your team dreads. In 45 minutes we’ll tell you whether an agent can clear it, what it has to touch, and what production actually takes. If the answer is no, we’ll say no.