@kryptosai/veros

Plug into any EHR. Ask questions in plain English. Get cited answers from the chart.
Every AI claim gets fact-checked before it reaches a clinician.

Quick demo

git clone https://github.com/KryptosAI/veros.git && cd veros && npm install && npm start
# → http://localhost:3100 (4 synthetic ophthalmology patients pre-loaded)

# Ask a question
curl -X POST localhost:3100/api/query/demo \
  -H 'Content-Type: application/json' \
  -d '{"question":"is this patient allergic to penicillin","patientId":"patient-002","userId":"user-dr-nguyen"}'

{
  "understanding": "Checking for allergies or adverse reactions",
  "answer": "YES — Maria E Santos has a documented Penicillin. HIGH RISK. Anaphylaxis with respiratory distress in 1990.",
  "citations": [{
    "sourceType": "AllergyIntolerance",
    "sourceId": "allergy-002",
    "display": "Penicillin — HIGH RISK (Confirmed)",
    "snippet": "Anaphylaxis with respiratory distress — avoid penicillin-class antibiotics."
  }],
  "policy": "all_cited"
}

# Verify an AI claim — decomposes compound claims into atomic propositions
curl -X POST localhost:3100/api/verify/demo \
  -H 'Content-Type: application/json' \
  -d '{"claim":"Patient has diabetic retinopathy, is on metformin, and is allergic to penicillin","patientId":"patient-001","userId":"user-dr-chen"}'

{
  "verdict": "PARTIALLY_VERIFIED",
  "reason": "2 verified, 0 contradicted, 1 unverifiable.",
  "decomposed": true,
  "propositions": [
    { "proposition": "Patient has diabetic retinopathy", "verdict": "VERIFIED", "reason": "Confirmed: 'Diabetic Retinopathy — Proliferative, Both Eyes' is documented (active)." },
    { "proposition": "is on metformin", "verdict": "VERIFIED", "reason": "Confirmed: Metformin 1000mg — active (order)." },
    { "proposition": "is allergic to penicillin", "verdict": "UNVERIFIABLE", "reason": "No allergy or medication history found for penicillin." }
  ]
}

# Generate a differential diagnosis
curl -X POST localhost:3100/api/differential/demo \
  -H 'Content-Type: application/json' \
  -d '{"symptoms":"62yo diabetic with progressive blurry vision OS x6mo, no pain, no flashes","patientId":"patient-001","userId":"user-dr-chen"}'

Rank	Diagnosis	Likelihood	Supporting
1	Diabetic Macular Edema Progression	high	HbA1c 8.2%, PDR, anti-VEGF therapy
2	Vitreous Hemorrhage	medium	PDR both eyes
3	Cataract	medium	Age 62, diabetic
4	Tractional Retinal Detachment	low	PDR both eyes
5	NAION	low	Age 62, hypertension

Each diagnosis includes suggested tests to order and questions to ask the patient.

What it does

AI systems are being embedded in every EHR. They write notes, suggest diagnoses, summarize charts. They also hallucinate. A fabricated allergy or a wrong medication in a clinical note isn't a typo — it's dangerous. Veros is a validation layer: submit a claim, and it checks whether the actual patient record backs it up, contradicts it, or has no data either way.

Features

🗣 Plain English queries	📋 Cited FHIR sources	🔍 AI claim verification
Ask anything about a patient. LLM interprets intent, searches FHIR, answers with citations.	Every answer links to the exact chart source — resource type, ID, author, date, clinical snippet.	Submit a claim. Get VERIFIED, CONTRADICTED, or UNVERIFIABLE with per-proposition evidence.
🩺 Differential diagnosis	🔒 RBAC + audit trail	🏥 FHIR-native · SMART auth
Describe symptoms. Get ranked differential with supporting/contradicting evidence, suggested tests, patient questions.	9 clinical roles. Per-resource access. Hash-chained tamper-evident audit log.	Speaks FHIR R4. Authenticates via SMART on FHIR. Works with Epic, Oracle, Medplum, OpenEMR, or anything with a FHIR API.

Install

npm install @kryptosai/veros

git clone https://github.com/KryptosAI/veros.git
cd veros && npm install
npm start          # → http://localhost:3100
npm run eval       # 27 unit tests
npm run benchmark  # citation precision/recall

Enable natural language understanding (otherwise falls back to deterministic regex):

export DEEPSEEK_API_KEY="sk-your-key"
npm start

Open source under MIT. PRs and issues welcome on GitHub.

License

MIT © KryptosAI