The operating network for the healthcare workforce.
Verify a clinician once. Reuse everywhere.
Healthcare doesn't have a staffing problem. It has a trust problem.
When a role opens, a clinician is re-verified, re-credentialed, and re-privileged from scratch — a ~78-day clock before they can work and bill. Then it runs again at the next facility. The verified work is non-portable; it never compounds.
Hired clinicians sit idle for months — facilities eat lost billable revenue and pay agency premiums to backfill.
Credentials expire on rolling clocks — one miss becomes a survey finding or a CMS clawback.
…then the whole clock runs again at the next facility.
Every facility rebuilds trust from zero, in its own silo.
Today's tools are a dozen point solutions stitched together — a job board, a credentialing inbox, a background-check vendor, a scheduler, a monitor nobody opens until something lapses. No memory between facilities; the same work, three times. Every handoff loses time.
Rōvn runs the whole path as one system — and makes the credential reusable.
Target: ~78 days to under 14. Our build target, not a delivered result.
One clinician verifies once. The same source-backed record clears them at the next facility — the worker carries it. Compounds with density.
One coherent system. Every decision provable. A human on every regulated call.
Agents interact and run as a single operator over a shared spine — not separate tools handing tasks off to each other.
Every determination is independently verifiable: a second institution can replay how it was decided, which primary sources it used, and whether anything was altered.
When confidence is low or the call is regulated, the system stops and routes to a named human.
Every clinician verified makes the network more valuable.
The network compounds as density accrues. Today the rail is built; the density is what the first pilots create — each verified clinician becomes reusable supply the next facility doesn't have to re-verify.
We occupy the one position incumbents can't take.
No incumbent can follow without convincing every customer to surrender their silo — a business-model reversal, not a feature. The Plaid-vs-banks inversion.
Every operated workflow produces outcome-labeled data that sharpens the system — an asset a point tool can't accumulate. It compounds with density, and turns on once we operate the first pilots.
Two things just became true at once.
Neither was true five years ago. The overlap is the opening.
Frontier reasoning over messy, inconsistent credentialing documents — under a BAA — became economically viable around 2024–25, not 2020. You could not have built a coherent operator before the models were ready.
That's a 2–3 year window, and it's open now.
Pre-launch, and clear about which rung we're on.
tier: VERIFIED
hash 7a3f…b1
sig ✓ Ed25519
tier: VERIFIED
prev 7a3f…b1
sig ✓ Ed25519
tier: VERIFIED
prev e9c2…4d
sig ✓ Ed25519
The verification layer is live in production today — the hard infrastructure others skip. What it has not yet touched is a real roster.
On a synthetic corpus. No paying customer yet; the 14-day number is a target.
Closing our first paying Operator Pilot — proving 78→14 on a live roster — is what this raise funds.
Builders and operators who have shipped in healthcare.
Built Rōvn's verification architecture and a real-time trading system. 3 yrs healthcare sales (Boehringer Ingelheim). Paused his other ventures to go all-in.
Built Rōvn's attestation pipeline as founding engineer. B.S. Cybersecurity (3.96 GPA); ran security/infra for 200+ users.
Rebuilt a healthcare data pipeline from >1 week to ~90 min across 15M+ records. ~6 yrs SWE, MS CS, agentic AI.
3.5 yrs at Athenahealth (EHR) + Commure (RCM) — the deepest healthcare-software résumé on the team.
Facilities pay per operated lifecycle. The network is the long game.
provider organizations & ASCs in band. Healthcare-workforce verification is a multi-billion-dollar, mostly-manual spend.
privileging → payer enrollment → two-sided network. Domain-general — teacher licensing, CDL/DOT, childcare, trades.
Rōvn becomes the operating network the entire healthcare workforce runs on — and the rail for all regulated labor.
Prove the ~78 days → under 14 target on live rosters.
Privileging, payer enrollment — the full application-to-billable operator.
Turn the rail into a living network that compounds with every pilot.
Three components make a determination portable.
Append-only and Ed25519-signed. Each entry links to the prior, so any edit breaks the chain. Public verify endpoints make the record tamper-evident.
Every fact carries its provenance, so the record is replayable, not asserted.
source url
timestamp
depth tier
validity window
Worker-owned and consented. The same verified claim is reusable across facilities, released under the holder's grant.
What makes a coherent autonomous operator trustworthy.
The NSF thesis. Four properties have to hold at once — and they have to hold across a live, regulated, multi-agent network, not a demo.
Every agent decision carries its inputs, its logic, and a source receipt. Nothing happens that can't be replayed and explained line by line.
The system knows when it must not decide. Low confidence or a regulated call routes to a human with the reasoning and tier-labeled facts attached.
Verified once, carried across facilities. A worker-owned record with depth labels and validity windows — proof that travels instead of being rebuilt.
Trust has a freshness clock. Sanctions, expirations, and status changes are re-checked on cadence — a cleared worker doesn't silently drift out of compliance.
Almost no one can make a coherent multi-agent system whose every decision is attestable, that knows when to defer, across a live regulated network. That combination is the moat.
The graph creates the model. The model does not create the graph.
commoditizing reasoning
outcome-labeled data
predict, never decide
the graph buys it
Outcome labels, not credentials — you only get them by operating real workflows. We rent the commoditizing reasoning and own the data no one else can get.
BloombergGPT cautionary tale: the data moat outlasts the model. The graph is the durable asset; the LLM is the option it buys — never a from-scratch pretrain.
43 roles × 51 jurisdictions × 36 primary sources = 2,193 cells.
The coverage matrix every credentialing claim resolves against. Each cell is a role / jurisdiction pair backed by a primary source — receipt-bound, depth-labeled, replayable.
Built for the 2026–2028 regulatory state.
Procurement-safe by construction. Every regulated decision stays human-owned, every fact carries a source receipt, and the audit trail is replayable on demand.
Claude runs under AWS Bedrock with a BAA. Only credential metadata moves through the AI gateway — no clinical PHI.
Every receipt links to the prior. S3 Object Lock, 7-year retention, replayable for JCAHO PSV or CMS recoupment defense.
SOC 2 in progress with Drata, not yet certified. NCQA CVO alignment underway, not certified. Stated honestly.
We don't launch as a network. We enter as a paid operator.
It pays from day one — a 90-day Operator Pilot at $12K.
Application to billable, plus roster monitoring — all inside the facility.
Every verification is a reusable credential — a byproduct of the work we're paid for.
A clinician shows up already-proven — the network effect compounds with density.
Independently-owned provider groups + ASCs, 30–1,000 clinicians; Southeast first. Too small for enterprise credentialing software, no staff to run it — the gap we fill.