ARDIA PRECISION HEALTHGoverned AI for healthcare revenue & precision care
What we solve

Focused agents for the work that
quietly drains labs and care teams.

Ardia builds narrow, governed AI agents across the areas we're focused on today — lab reimbursement (toxicology & molecular), precision medicine & genomics, the PAMA rate cliff, respiratory & pulmonary care, and elderly care — each grounded in the same de-identification and evaluation kernel, and each mapped to a delivery phase.

Pre-revenue · seed stage Governed by design Not a diagnosis
Segment 01

Independent Labs · Reimbursement

Phase 1 · building now
3 solutions
Full transparency

Every solution, in 360°

Eighteen dimensions per model — problem, buyer, architecture, data flow, standards, evidence, evaluation design, regulatory and liability posture, price and margin, competition, limitations and roadmap. Including what is not built.

Open the 360° views →
Segment 02

Precision Medicine & Health

Phase 1–2 · genomics & PGx reimbursement
4 pillars · MolecuIQ

Personalise each patient's prevention and treatment using their genomic and molecular profile — the right drug and dose, therapy matched to the molecular target, inherited risk caught early. Ardia builds the decision-support (roadmap) and the access layer (MolecuIQ, today) that gets the enabling tests covered. Non-diagnostic.

Segment 03

Chronic & Respiratory

Phase 3 · 2027–28 · reimbursement lane; prediction FDA-gated
1 agent
Segment 04

Elderly Care

Phase 1 build · pilot under BAA · 2027
Aria + 9 modules
Segment 05

Platform · Governance

Foundational · inherited by every model
2 kernels
Solution blueprint · how we build it

Lab Reimbursement — the blueprint

Policy-grounded, fully cited denial appeals and pre-submission necessity checks for independent toxicology and molecular labs — every draft is written for a human biller to review and submit, never auto-filed.

Four pillars — what it is & how it's built
🧪
ToxIQ — toxicology necessity & appeals

ToxIQ reviews definitive (G0480–G0483) and presumptive (80305–80307) drug tests pre-bill and drafts post-denial appeals. A Claude persona under guardrails, grounded on NCCI PTP edits + MUEs and CARC/RARC pairs (CO-50 necessity, CO-97 bundling), plus a non-diagnostic CPIC-informed PGx flag on metabolizer status.

🧬
MolecuIQ — molecular & genomic appeals

MolecuIQ drafts appeals for denied molecular claims — NGS tumor panels (81445, 81455), AMA PLA codes, and single-gene tests. A Claude persona under guardrails grounded on MolDX policy, Palmetto GBA Z-codes (DEX registry), and Local Coverage Articles, it builds medical-necessity CO-50 appeals mapping the ordered test to the payer's covered indication.

📚
Cited policy retrieval + code validators

Every appeal sentence anchors to a current source. A retrieval layer indexes CMS CLFS fee schedules, NCCI edit tables, MolDX Local Coverage Articles, and payer LCDs; RAG attaches an inline citation to each claim, while a deterministic validator checks CPT/HCPCS codes and NCCI edit pairs before any narrative is generated — no uncited assertions.

🧑‍⚕️
Human-in-the-loop, non-diagnostic handoff

Nothing is auto-filed. Each run yields a review-ready packet — denial root cause, cited policy, corrected-coding suggestion — that a biller edits and submits. Structured as HL7 FHIR resources (Claim, ClaimResponse, DocumentReference) for LIS/RCM handoff, with a per-draft audit log; output is non-diagnostic and pending human sign-off.

Models & engines
ToxIQ Modelled target
Claude persona under guardrails

Toxicology pre-submission necessity checks and post-denial appeals for definitive/presumptive drug testing

MolecuIQ Modelled target
Claude persona under guardrails

Drafts appeals for denied molecular/genomic claims — NGS panels, PLA codes, and MolDX-governed tests

Coding & edit validator In build
deterministic engine

Checks CPT/HCPCS coding and NCCI PTP/MUE edit pairs before any appeal narrative is drafted

Frameworks & standards we build on
AMA CPT & CMS HCPCS Level II codes (G0480–G0483, 80305–80307, 81445, 81455, PLA)CMS Clinical Laboratory Fee Schedule (CLFS) & PAMANCCI — PTP edits and Medically Unlikely Edits (MUEs)MolDX Program / Palmetto GBA DEX Z-code registryCMS Local Coverage Determinations & Articles (LCD/LCA)X12 CARC/RARC denial codes (835/837 EDI)CPIC pharmacogenomics guidelines (PGx overlay)HL7 FHIR interoperability standard
Phase roadmap — what's now, next & later
Phase 1 — BuildNow
2026

Ground ToxIQ and MolecuIQ on the CMS/MolDX policy corpus, build the deterministic CPT/NCCI edit validators, and generate cited draft appeals on synthetic and public de-identified denial scenarios — no PHI.

Phase 2 — Pilot under BAANext
2027

Pilot with a design-partner independent lab under a signed BAA on limited/de-identified data; billers review and edit every draft; qualitatively assess citation accuracy and edit-acceptance.

Phase 3 — ScaleLater
2027–2028

Broaden payer LCA/LCD coverage and PLA-code breadth, wire FHIR appeal packets into LIS/RCM systems, and extend beyond toxicology and molecular to adjacent lab-reimbursement lines.

Honest limitation: ToxIQ and MolecuIQ are LLM personas with no published accuracy; every appeal is policy-cited for a human biller to verify and submit — never auto-filed, never diagnostic, with no reimbursement outcome promised. Ardia is pre-revenue with no lab clients or billing results yet.
Where we are. Ardia is pre-revenue and seed-stage. The agents above describe intended capability and work in progress — not deployed clinical outcomes, production results, or customers. Sentinel (de-identification) and Crucible (release-evaluation) are the governance kernels we are building; their test suites are a target in progress, not a measured result yet. The one model with a measured benchmark today is Cadence — 95.45% held-out on public UCI HAR (subject-independent). Nothing here is a diagnosis or medical advice.