Same note. Same codes. Every time. Zero PHI exposure. The clinical extraction logic that doesn't exist anywhere you can buy it.
The note is the source of truth. But between the note and the claim, there's a gap — manual coding, weak auto-suggest, or nothing at all. Everything downstream inherits that gap.
Unstructured narrative in EHR (Epic, Cerner, etc.)
Manual coder, weak EHR auto-suggest, or no coder at all.
Missed specificity, wrong codes, missed HCCs, vague documentation
Scrubbing + edits catch format errors — but can’t fix bad codes
15% denied. $25–40 rework per denial.
Yes. One API endpoint. No workflow changes. PHI never leaves your network.
Stage 1 runs on-device (entity extraction, PHI containment). Stages 2–4 run on our cloud via API. Your data stays in your network. Our IP stays in ours.
POST a PHI-free intermediate representation, receive structured claim-ready JSON. Already built, tested, deployed on Google Cloud Run.
Epic, Cerner, fax, OCR, dictation, wall-of-text. The engine normalizes and extracts from whatever you feed it. No EHR dependency.
Runs in parallel with your existing pipeline. Compare our output against what you currently receive. Zero disruption to production.
Under 2 weeks to pilot. We ship a thin client for Stage 1. You point it at notes. Results flow through our API. You compare against your current output.
Every output is traceable, reproducible, and evidence-backed. No black boxes. No hallucinated codes. No PHI exposure.
Same note, same codes, every time. Output verified with SHA-256 hash comparison across runs. No model drift, no retraining instability, no stochastic variation.
Three-layer fail-closed egress gate. Clinical notes are structurally transformed into SNOMED concept documents — not redacted, not masked. The rebuilt document preserves clinical meaning with zero patient identifiers.
Every code has: the sentence that supports it, the assertion proof, the family resolver path, the suppression evidence, and the tier classification reason. Auditors can trace any output back to the note.
90%+ of output is deterministic rules. The AI layer (∼10%) can only propose — it cannot change billing. Every AI suggestion is validated by the engine before surfacing. The AI never sees PHI.
“Run the same note 100 times. Get the same codes 100 times. No AI-first system can make that claim.”
43 core modules. 5 pipeline stages. 2+ years of clinical engineering. CMS publishes codes. SNOMED publishes the ontology. The decision logic that connects them doesn't exist anywhere you can buy it.
Semantic-tag gated SNOMED index, abbreviation expansion, section-aware matching, 6-signal candidate scoring
Multi-axis code resolution for every major ICD-10 family. Heart failure alone has 12 codes across 2 axes with HCC boundary guards
Structural transformation (not redaction) that preserves clinical meaning. Three-layer fail-closed egress gate
4.4M SNOMED relationships wired into O(1) query API. 851 explained-by pairs, acuity variants, complication chains
Per-entity evidence annotation, 5-tier classification, entity dossier builder, assertion detection operator algebra
255 integration tests, 19 golden tests, blind evaluation, reproducibility verification, adversarial audit
2–3 years · 3–4 engineers + clinical informaticist
And that assumes you know what to build. The clinical edge cases — dictated notes, assertion scoping, family axis resolution, HCC hierarchy conflicts — are discovered through thousands of test notes, not designed upfront.
You own Iodine (during encounter) and the claim lifecycle (AltitudeAI). The extraction layer between them is the one piece you don't control.
Iodine (Yours) — CDI. Improves documentation while the doctor writes.
VMC (The Gap) — Turns the finished note into claim-ready codes. Deterministic. Any input source.
Waystar Platform (Yours) — Scrubbing, submission, denial management, appeals. 6B transactions/year.
A competitor that was years behind now has 43 production modules, 50+ family resolvers, and a PHI-safe AI architecture — deployed.
They get an Electron app, a cloud API, and a tested pipeline. They can offer extraction to their clients in weeks, not years.
574K concept mappings, 851 suppression rules, 4.4M graph relationships. The decision logic that took 2+ years to build walks out the door.
Right now, no one in the market knows the extraction layer is a solvable problem. A competitor who acquires this doesn’t sell it as a product — they embed it. Every garbage note across their entire platform starts producing better data. Their denials drop. Their first-pass rates climb. And you’re still feeding AltitudeAI the same incomplete input.
Iodine improves the note. VMC extracts the codes. Waystar processes the claim. End-to-end, one platform.
Extraction quality feeds directly into denial prevention, charge integrity, and appeals. Cleaner input, fewer denials.
Platform-level deployment. Every Waystar client gets better first-pass acceptance without buying a separate tool.
You own all three layers. The only company with CDI + extraction + claim lifecycle under one roof.
Industry average: 15% initial denial rate. $25–40 rework cost per denial. A 5% denial reduction across 30,000 clients is measurable in the first 30 days.
A 30-day parallel run on notes already flowing into your platform.
Our extraction alongside whatever you currently receive.
You see the delta. No commitment until you see data.