SOLUTION · HEALTHCARE · ENTERPRISE AI AGENTS ● PRODUCTION

Your Medical Supply Ordering,
run autonomously.

from $0.99 per fulfilled healthcare mission From referral to billing status: accepted — without a human in the loop. Four specialist agents running in production today.

Request a Demo Meet the Fleet See the Workflow

The most document-heavy supply chain in healthcare.

When a home health clinician generates a medical supply referral in Order Intake, it triggers a workflow that should be routine — and rarely is. Required clinical documents must be collected from providers across multiple formats and systems: Face-to-Face Encounter Forms, Standard Written Orders, Assignments of Benefit, Title XIX Medicaid forms. Each carries its own regulatory requirements. Each must be validated against HCPCS codes, ICD-10 diagnoses, physician NPI numbers, and payer-specific rules before a single item can ship.

Prior authorization requirements change by product, by payer, and by patient. A single missing physician signature on one document can stall the entire order. When prior authorization affects only a subset of line items, the order must be split — clean items submitted immediately, authorized items held until approval is confirmed. Payer responses are asynchronous. Callbacks arrive on their own schedule.

Today, human operators run this process manually. They interpret Order Intake XML outputs. They email providers and wait for responses. They re-contact physicians when signatures are missing or diagnosis codes are incorrect. They submit EDI transactions by hand, parse the responses, and route exceptions to the right teams. They track order splits across systems that were never designed to talk to each other.

The bottleneck is not intelligence. Every step in this workflow is rule-based, document-driven, and well-defined. What has been missing is execution infrastructure capable of sustaining a stateful, multi-step, multi-party process reliably across the failures and delays that characterize real clinical environments. PLRX eliminates that friction — and the Medical Supply Ordering workflow is where we proved it.

Eight steps. Zero humans.

Every step below is executed autonomously by the PLRX agent fleet. The PLRX Durable State Machine persists workflow state across all of them — a mission that pauses at step 2 for three weeks while awaiting provider documents resumes at exactly the right state, with all previous context intact, when documents arrive.

  1. 01
    Ingest & Record
    The Order Intake system sends an XML referral to the PLRX Order Connector. The connector normalizes it into a structured order payload and publishes it to the event stream. The Customer Service Agent records the inbound A2A request, persists the Phase 1 order to its data store, and starts a PLRX Durable State Machine workflow. Mission ID assigned. State: MISSION_RECEIVED.
    XML/HTTPSrecord_inbound_a2a_requestpersist_phase1_order
  2. 02
    Phase 1 Validation — Parallel
    Two validations run in parallel inside the durable workflow. NPI validation: the physician's National Provider Identifier is submitted to the NPI Registry and validated via the Luhn algorithm. Payer eligibility: the Billing Service Agent builds and submits an EDI 270 eligibility request to the clearinghouse and parses the EDI 271 response to confirm active patient coverage. Both must pass before the mission advances.
    validate_physician_npicheck_phase1_payer_eligibilityparallel executionEDI 270/271
  3. 03
    Document Collection & Validation
    The Clinical Documentation Specialist sends a structured document request to the referring provider via email, listing all required documents by type. As documents arrive, each is processed through the multimodal OCR pipeline for text extraction, then validated against regulatory requirements: HCPCS codes, ICD-10 diagnosis codes, physician NPI numbers, and signature presence. A Levenshtein-distance algorithm auto-corrects common ICD-10 transcription errors. Invalid documents trigger an immediate correction request to the provider. Missing documents trigger a 48-hour reminder loop. The workflow suspends durably — input_required — until all documents pass. There is no timeout. The mission waits as long as necessary.
    document_collectiondocument_follow_upOCR extractionLevenshtein algoNPI Registry
  4. 04
    ML Denial-Risk Scoring
    Once all documents are validated, the Data Scientist Agent submits the complete order to the ML inference service. A gradient boosting model trained on historical order outcomes scores the order for denial risk — checking for over-allowable quantities, clinical inconsistencies between diagnosis codes and HCPCS codes, and payer-specific flags. The agent returns a confidence score, a set of risk flags, and a recommended action: proceed, split, or escalate. No LLM is involved in the statistical scoring path.
    denial_risk_scoringML inferencedenial risk flagsno LLM in path
  5. 05
    Prior Authorization Submission
    The Billing Service Agent identifies which HCPCS codes require prior authorization for the patient's payer. For each code requiring auth, it submits an EDI 278 prior authorization request to the payer's FHIR R4 endpoint. Payer responses are asynchronous — the workflow suspends durably until each callback is received. Authorization decisions (approved, denied, modified) are captured and passed to the orchestrator.
    prior_auth_submissionEDI 278FHIR R4async callback
  6. 06
    Order Splitting (When Required)
    When prior authorization is required for only a subset of line items, the Customer Service Agent splits the order. Clean items — those with active eligibility and no prior auth requirement — are submitted to the billing system immediately. Prior-auth items are held in a separate sub-workflow until authorization is confirmed. Each sub-order tracks its own state independently inside the PLRX Durable State Machine. Both halves are reconciled into a single mission record on completion.
    order split logicparallel sub-workflowsclean vs prior-auth items
  7. 07
    Submission to the billing system
    The validated, authorized order — or each split sub-order — is submitted to the order fulfilment system medical fulfilment platform via the order fulfilment system adapter. The Customer Service Agent tracks the fulfilment status, monitoring the order fulfilment system callback for acceptance (status 19) or rejection. On rejection, the mission enters an exception state and the relevant exception-handling agent is notified. On acceptance, the mission completes.
    order fulfilment system adapterstatus 19 · acceptedfulfilment tracking
  8. 08
    Mission Complete
    The PLRX Durable State Machine transitions to MISSION_COMPLETED. All state transitions, document events, validation outcomes, EDI transactions, model inference calls, and AI prompts are captured in the WORM audit log. The mission record is closed. The customer is billed for one settled outcome.
    MISSION_COMPLETEDWORM audit logfrom $0.99 per settled mission

Four agents. One mission.

Each specialist agent is an independently deployable service owning its own durable workflow instance, task queue, and data schema. The orchestrator delegates to specialists via A2A — it never reads a specialist's data directly. Specialists execute in parallel where the workflow permits. Each can be updated, scaled, or restarted without affecting the others.

The cost share column reflects live production allocation from Mission Control: the Customer Service Agent accounts for the largest share because it owns the full mission lifecycle. The Data Scientist Agent is the lowest because ML inference, while critical, is computationally lean.

The real production workflow.

Every state shown below is taken directly from the live production execution. The parallel Phase 1 validation, the input_required suspension during document collection, the ML scoring step, the FHIR R4 prior auth submission — all real. The activity names shown are the actual function names executed by the PLRX Durable State Machine in production.

PRODUCTION OBSERVATIONS
parallel exec NPI validation and payer eligibility run simultaneously in Phase 1 — both durable independently
C2 retry map_phase2_order retried twice in observed executions — recovered cleanly, no state loss, no duplicate side effects
longest wait Missions suspended for multi-week provider document collection resume correctly — zero state drift
auto-correct ICD-10 code C89.314 → L89.314 corrected automatically via Levenshtein distance — mission proceeds without provider re-contact

Pay per settled mission.
Nothing else.

From $0.99
per settled medical supply mission

A settled mission is one that reaches order fulfilment system fulfilment status 19 — accepted by the healthcare supplier after passing through the full agent workflow. You are not billed for compute time, token consumption, API calls, or workflow duration.

A mission that waits three weeks for a physician to return a signed form does not accrue charges during the wait. A mission that fails validation or is rejected by order fulfilment system is not billed at all.

  • Accepted by order fulfilment system — mission billed once at from $0.99
  • No charge for workflow duration or waiting time
  • No charge for token consumption or API calls
  • Failed or rejected missions not billed
  • No setup fees · No monthly minimums
  • No infrastructure or compute charges

Your Medical Supply Ordering orders, running autonomously.

The Medical Supply Ordering fleet is in production today. We can connect it to your Order Intake instance and show you a live mission in your environment.

Request a demo and we will walk you through the full workflow — document to fulfilment.

Request a Demo Next: Revenue Cycle →