Developer documentation

Introduction to NeuralCare AI

NeuralCare AI is a GPU-accelerated clinical decision intelligence platform that runs deep learning inference on electronic health record data to surface early disease risk signals for clinical teams. The platform processes patient data in real time — average inference latency of 47ms — and integrates with existing EHR systems via FHIR R4 and HL7 v2.x standards.

The system is built on a transformer architecture trained across 2.1 billion de-identified patient records using federated learning. No raw patient data ever leaves the institutional boundary during training or inference aggregation.

Quick Start

Get your first inference in under 5 minutes. Install the Python or Node.js SDK, then run a patient risk assessment with a single API call.

Install the SDK

# Install
pip install neuralcare

# Quick inference
from neuralcare import NeuralCareClient

client = NeuralCareClient(api_key="nc_live_xxxxxxxxxxxx")

result = client.infer(
    patient_id="pt-928374",
    model="oncology-v4",
    fhir_bundle=patient_fhir_bundle
)

print(result.risk_score)   # 0.847
print(result.signals)      # ['CBC_trend', 'fatigue_pattern']
print(result.latency_ms)   # 43

// Install
npm install @neuralcare/sdk

// Quick inference
import { NeuralCareClient } from '@neuralcare/sdk';

const client = new NeuralCareClient({ apiKey: 'nc_live_xxxx' });

const result = await client.infer({
  patientId: 'pt-928374',
  model: 'oncology-v4',
  fhirBundle: patientFhirBundle
});

console.log(result.riskScore);  // 0.847
console.log(result.latencyMs);  // 43

curl -X POST https://api.neuralcareai.tech/v1/infer   -H "Authorization: Bearer nc_live_xxxx"   -H "Content-Type: application/json"   -d '{"patient_id":"pt-928374","model":"oncology-v4"}'

# Response
{
  "risk_score": 0.847,
  "latency_ms": 43,
  "model_version": "oncology-v4.2.1"
}

FHIR R4 Integration

NeuralCare consumes FHIR R4 bundles directly. The platform accepts Patient, Observation, Condition, MedicationRequest, and Procedure resources. Unsupported resource types are silently skipped.

Supported EHR Connectors

Federated Learning

NeuralCare supports optional federated participation — your institution's data improves the global model without ever leaving your premises. Gradient updates are differentially private and cannot be used to reconstruct any patient record.