Same note. Same codes. Every time. Zero PHI exposure. The clinical extraction logic that doesn't exist anywhere you can buy it.
Projected VMC Impact on Waystar Platform
The note is the source of truth. Between the note and the claim, there's a gap. Everything downstream inherits it.
Iodine (yours) improves documentation during encounter
Manual coder, weak auto-suggest, or nothing.
AltitudeAI catches format errors & payer rules — can’t fix wrong codes
Scrubbing + edits applied. Bad codes still pass through.
15% denied. $25–40 rework per denial.
One API endpoint. No workflow changes. PHI never leaves your network.
Stage 1 on-device (extraction, PHI containment). Stages 2–4 cloud via API. Your data stays in your network.
POST PHI-free IR, receive structured claim-ready JSON. Built, tested, deployed on Google Cloud Run.
Epic, Cerner, fax, OCR, dictation. No EHR dependency.
Parallel with existing pipeline. Compare output. Zero disruption.
Under 2 weeks to pilot. Thin client for Stage 1. Point at notes. Compare against current output.
Every output is traceable, reproducible, evidence-backed. No black boxes. No hallucinated codes.
Same note, same codes, every time. SHA-256 hash verified. No model drift.
Three-layer fail-closed egress gate. Structural transformation, not redaction. Zero patient identifiers.
Every code: supporting sentence, assertion proof, resolver path, suppression evidence, tier reason. Auditors trace any output to the note.
90%+ deterministic rules. AI (~10%) can only propose. Engine validates before surfacing. 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 modules. 5 pipeline stages. 2+ years. The decision logic connecting CMS codes to SNOMED ontology 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: 12 codes, 2 axes, HCC boundary guards. Pattern is simple once discovered — discovering it requires clinical coding + architecture expertise in the same person.
Structural transformation preserving clinical meaning. Three-layer fail-closed egress gate
4.4M SNOMED relationships, O(1) query API. 851 explained-by pairs, acuity variants, complication chains
Per-entity evidence annotation, 5-tier classification, entity dossier, assertion operator algebra
255 integration tests, 19 golden tests, blind eval, reproducibility verification, adversarial audit
2–3 years · 3–4 engineers + clinical informaticist
Clinical edge cases are discovered through thousands of test notes, not designed upfront.
You own Iodine (during encounter) and 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 finished note into claim-ready codes. Deterministic. Any input.
Waystar Platform (Yours) — Scrubbing, submission, denial management, appeals. 5B+ transactions/year.
A competitor gets 43 production modules, 50+ family resolvers, PHI-safe AI architecture — deployed.
Electron app, cloud API, tested pipeline. They offer extraction to their clients in weeks, not years.
574K mappings, 851 suppression rules, 4.4M graph relationships. 2+ years of decision logic walks out.
No one knows the extraction layer is solvable. A competitor embeds it — every garbage note produces better data. Their denials drop, first-pass rates climb. You’re still feeding AltitudeAI the same incomplete input.
Iodine → VMC → Waystar. End-to-end, one platform.
Cleaner input → fewer denials, better charge integrity, stronger appeals.
Platform-level deployment. Better first-pass acceptance without a separate tool.
You own all three layers. Only company with CDI + extraction + claim lifecycle.
Conservative modeling. 50.0M annual notes — a fraction of 5B+ platform transactions. All inputs from public filings and industry benchmarks.
Throughput = capacity expansion, not headcount reduction. Conservative modeling. Framework for diligence.
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.