Physician-built clinical NLP
The glass box for clinical language.
Archimedes Medical turns unstructured clinical notes into SNOMED CT–anchored, deterministic, auditable intelligence — plain-English clarity for patients, and proof-grade structure for the platforms that serve them.
72 y/o M s/p L fem-pop bypass, c/o rest pain L foot. PMH: DM2, HTN, CKD 3, HFrEF (EF 30%). On metformin, lisinopril. Denies chest pain negated. Abdomen soft normal.
What “DM2” means
Type 2 diabetes
A long-lasting condition where your body doesn’t use insulin well, so the sugar in your blood stays too high. Over time it can affect your heart, kidneys, nerves and blood vessels.
Written at a 7th-grade reading level · one of 264,000+ plain-English definitions
Medication audit · grounded CDS
Relative precaution: metformin × CKD stage 3
Confirm current eGFR; reassess dosing as renal function declines. Labeling advises against starting metformin at eGFR 30–45 and contraindicates it below 30.
CDI specificity nudge
HFrEF is documented separately. Ask the physician to clarify the link — the suggestion is a SNOMED descendant of what they wrote, mapped through the NLM ICD-10 crosswalk, with the RAF delta shown. Never an LLM guess.
550K+
clinical embeddings in the SNOMED resolver
264K+
plain-English concept definitions
>97% F1
geometric trust gate on concept matches
10M
PubMed & StatPearls articles, SNOMED-indexed
0
bytes of PHI persisted — stateless, in-memory API
One engine, two doors
The same engine that explains a CT report to a worried parent can defend a code in an audit.
For patients & families · free
RosettaMD
Paste or upload a lab result, imaging report or discharge summary. Every medical term lights up with a plain-English explanation — in context, not from a glossary. Nearly 9 in 10 U.S. adults struggle to understand health information; this is for them.
- No account, no ads, no tracking
- Web app and Chrome extension — works on PDFs and patient portals
- Your document is never stored
For health-tech platforms · B2B API
RosettaMD-AI
The “Intel Inside” for ambient scribes, EHR overlays and RCM tools. Send a draft note as JSON; get back SNOMED-resolved concepts, compliant HCC nudges, ranked literature and grounded decision support — rendered natively in your own UI.
- Stateless JSON-in / JSON-out REST — integrate in weeks
- Zero-retention by design shortens security review
- Usage-based pricing per encounter, with volume tiers
How the engine works
From messy note to verifiable answer in six deterministic stages.
Catch everything first, then prove every match. Patient notes and the literature are normalized into the same SNOMED CT space, so they can be compared concept for concept.
01 · Recall-first NER
Nested span detection tuned to miss nothing — clinical shorthand like s/p, c/o and lap-chole included.
02 · SNOMED resolution
A two-tier gate: near-exact matches pass; everything else must fall inside the concept’s calibrated Radius of Trust.
03 · Ontological filtering
Negated and normal findings are dropped. What remains is weighted by semantic type and rarity (concept-level BM25).
04 · Vector retrieval
A weighted concept vector pulls the top ~300 candidates from a Milvus index of PubMed, StatPearls and MIMIC-IV.
05 · OCMD re-ranking
Ontological Concept Mover’s Distance scores coverage of the patient and penalizes off-topic documents — only along valid SNOMED paths.
06 · Grounded output
Codes from IS_A filters and NLM crosswalks. CDS as schema-validated JSON citing only the patient and retrieved papers.
Glass box, not black box
Six rules the engine never breaks.
Compliance teams don’t trust probabilities. They trust a chain of evidence they can follow. Every output traces back to an ontology path, a crosswalk entry or a document ID.
RosettaMD-AI modules
Turn on what your roadmap needs.
A
Real-time CDI & HCC engine
Catches non-specific documentation before the note is signed and offers compliant specificity nudges with the RAF impact — hard-dollar ROI at the point of care.
For CFOs and revenue-cycle leaders
B
Evidence grounding & retrieval
Send a whole patient presentation, get back the most relevant literature in under 300 ms — ready to feed your own HIPAA-compliant LLM without building the search stack.
For CTOs and engineering leads
C
Audit-proof decision support
Three workflows — diagnostic gaps, medication contraindication audits, and adverse-event / prescribing-cascade checks — each with dual citations to the chart and the literature.
For CMIOs and quality officers
POST /v1/analyze → 200 OK
{
"query_type": "MEDICATION_AUDIT",
"negative_rule_triggered": false,
"findings": [{
"category": "RELATIVE_PRECAUTION",
"item_name": "metformin",
"patient_evidence": ["PMH: CKD 3"],
"literature_evidence": [{
"pmid": "verified",
"title": "…"
}],
"action_item": "Confirm eGFR; adjust"
}],
"cdi_nudges": [{
"from": "I10",
"to": "I11.0",
"path": "IS_A descendant",
"raf_delta": "+"
}]
}
Illustrative response shape
See the constellation
Every chart is a constellation of concepts.
Our DocVec Explorer projects each document’s SNOMED concepts into an interactive 3D map. Overlapping regions reveal shared disease mechanisms; disjoint regions surface unique comorbidities — so a clinician can see why a paper matched, not just that it did.
- Query the literature with an entire discharge summary
- Rare vascular phenotypes matched by shape, not keywords
- Cohort discovery for population-health research
Roadmap
Health literacy today. Clinical intelligence next.
Live now
- RosettaMD web app
- RosettaMD Chrome extension
- Semantic vocabulary search API
- DocVec Explorer demo (PubMed · StatPearls · MIMIC-IV)
In development
- Nested-span NER for higher recall
- Audited, re-calibrated SNOMED embedding table
- Bidirectional OCMD literature ranking
- Real-time CDI nudges with RAF scoring
Next
- SMART on FHIR decision-support panel for Epic & Cerner
- Design-partner pilots with ambient-scribe vendors
- Population health & patient digital twins
Built from the exam room
Physician-led. Privacy-first. Purpose-driven.
Archimedes Medical was founded by Dr. Chris Wixon, a practicing vascular surgeon in South Carolina’s Lowcountry. After years of watching patients arrive with folders of records they couldn’t read — and colleagues drowning in documentation — he set out to build an engine that understands clinical language the way clinicians do, and proves every step.
Meet the founder →“RosettaMD helped me read my own radiology reports with confidence.”
“It gave me answers when my health system didn’t.”
Bring audit-proof intelligence into your product.
We’re selecting design partners among ambient-scribe, EHR-overlay and revenue-cycle platforms. Tell us what you’re building.
Design-partner inquiries
Tell us about your platform, the module you're interested in (CDI & HCC, evidence retrieval, decision support, or an Explorer demo) and your timeline.
Request API accessRosettaMD explains medical language; it does not give medical advice. Clinical decision-support output is for clinician review only. SNOMED CT is a registered trademark of SNOMED International.