ARDIA PRECISION HEALTHGoverned AI for healthcare revenue & precision care
Back to Home
Technical Architecture

Clinical Intelligence Engine
Architecture

A HIPAA-ready AI platform built on Google Cloud Platform (Vertex AI, Cloud Healthcare API) with multi-layer clinical reasoning, payer-rules intelligence, and FHIR-native EHR integration.

Target architecture — pre-revenue. The layers below describe the system as designed. Live today: static hosting (Vercel) + Anthropic Claude / Google Gemini as the reasoning engine, plus the tested Meridian (PAMA math), Sentinel (de-identification) and Crucible (guardrails) components. Everything else is in build unless explicitly marked measured.
Layer 1 · Data Ingestion & Sources
📄

EDI Claims Data

ANSI X12 835/837 denial files. Planned real-time clearinghouse API connections. Bulk historical upload.

835 ERA837P/IReal-time
🏥

EHR Clinical Data

SMART-on-FHIR R4 API for physician notes, Rx history, lab orders, diagnoses.

FHIR R4EpicCerner
🧬

Genomic & Lab Results

HL7 v2.x LIS interfaces. VCF/FASTQ genomic data parsing. Structured result ingestion.

HL7 v2VCFORU
💊

Pharmacy & PBM Data

NCPDP SCRIPT prescription data. PBM formulary feeds. Drug-gene interaction database.

NCPDPRxNormSCRIPT
Layer 2 · Processing & Normalization
⚙️

Data Normalization Engine

Multi-format parser normalizing EDI, HL7, FHIR, and proprietary formats into a unified clinical-financial data model. Deduplication, entity resolution, and temporal alignment.

Mirth ConnectFHIR MapperNLP Pre-processing
🗂️

Clinical Document Analyzer

NLP-powered extraction of structured data from unstructured physician notes. Entity recognition for diagnoses, medications, procedures, and clinical findings with medical ontology mapping.

SNOMED CTLOINCRxNormICD-10
🔗

Patient Identity Resolution

Master Patient Index with probabilistic matching across payer, EHR, and lab systems. EMPI integration. HIPAA-safe tokenization for analytics.

EMPITokenizationMPI
Layer 3 · AI Clinical Reasoning Core (The Differentiator)
🤖

Claude AI Engine

Anthropic Claude API powers clinical reasoning. Healthcare-context prompting with payer-specific LCD rule injection. Mechanistic explanation generation.

Claude APIAnthropicRAG
📚

Payer Rules Knowledge Base

Planned — 400+ commercial payer policies, all 7 MAC LCD/NCA databases, and MolDX dossier requirements, refreshed from CMS feeds. Pre-revenue; in build.

7 MACsMolDX400+ Payers
🧬

Genomic Intelligence Module

Planned — multi-omics interpretation, PGx CYP metabolism modelling, biomarker-to-coverage mapping, and clinical-trial eligibility matching. Pre-revenue; in build.

CYP2D6/2C19NGSPGx
📖

Clinical Guidelines Engine

NCCN, ASCO, AHA, ACC, and 40+ specialty society guidelines mapped to payer coverage policies. Planned: Real-time citation generation for evidence-based rebuttals.

NCCNASCOAHA
Layer 4 · Output & Workflow
✍️

Recovery Narrative Generator

Payer-formatted appeal letters with clinical reasoning, evidence citations, and structured attestation. Generated in <90 seconds per claim — modelled target vs 2–4 hrs manual.

🔄

Workflow Orchestrator

Configurable reviewer queues, reviewer-approval rules, deadline tracking, and escalation workflows. Runs on Cloud Run.

📊

Revenue Analytics Module

Recovery tracking, ROI modelling, denial-pattern analysis, and payer performance benchmarking.

🔌

Clearinghouse Submission

Direct integration with major clearinghouses for electronic appeal submission. Paper submission PDF generation for payers requiring it.

Layer 5 · Infrastructure & Platform
☁️

Cloud Healthcare API & Healthcare Data Engine

HIPAA-eligible GCP services (BAA-ready). FHIR R4 store native APIs. US data residency in us-central1.

Cloud Healthcare APIHealthcare Data EngineFHIR R4
🌐

Google Cloud Platform

GCP Healthcare API. BigQuery for analytics. Vertex AI for custom ML models. HIPAA-eligible services.

GCPBigQueryVertex AI
🔐

Security Stack

AES-256 at rest. TLS 1.3 in transit. Zero-trust network. SOC 2 Type II target (Month 18). HITRUST roadmap.

Zero TrustSOC 2 (planned)HITRUST
⚛️

Upgrade-Ready Layer (Phase 4)

Architecture designed for quantum-classical hybrid computing. Post-quantum cryptography. Q2 2032+ target for quantum ML model integration.

Post-Quantum CryptoQML Ready2032+

Claim-to-Recovery in 6 Steps

1
Denied Claim Received
835 ERA or direct API delivers denial with reason code, payer, claim details, and service line breakdown.
2
Rules Engine Routing
Payer identified → MAC jurisdiction → applicable LCD/NCA retrieved → denial code classification established.
3
Clinical Context Retrieval
FHIR query pulls clinical documents — notes, Rx, labs, genomic reports — scoped to the patient and clinical question.
4
AI Reasoning & Gap Analysis
Claude API reasons documentation against LCD medical necessity criteria. Identifies specific gaps. Selects optimal evidence from clinical guidelines.
5
Narrative & Appeal Assembly
Structured appeal letter generated. Evidence package assembled. Confidence score calculated. High-confidence appeals queued for licensed reviewer approval before submission.
6
Submit & Track Outcomes
Appeal submitted via clearinghouse. Outcome tracked on ERA receipt. Model learns from wins. ROI calculated and reported.
✓ Recovery Complete

Built on Best-in-Class Healthcare Infrastructure

AI & Machine Learning

Google Gemini API — primary reasoning engine (Claude evaluated as a secondary provider)
Google Vertex AI — custom ML models & fine-tuning
RAG (Retrieval-Augmented Generation) — payer policy context injection
spaCy + Hugging Face — clinical NLP pre-processing

Healthcare Interoperability

HL7 FHIR R4 — EHR clinical data integration
ANSI X12 835/837 — claims and remittance EDI
SMART-on-FHIR — EHR app launch & authorization
Mirth Connect — HL7 v2 message transformation
NCPDP SCRIPT — pharmacy data exchange

Cloud & Data

Google Cloud Platform (GCP) — primary platform (us-central1)
BigQuery — analytics & reporting data warehouse
GCP Cloud Healthcare API — FHIR R4-native data store
Cloud Run — serverless microservices
Salesforce Health Cloud — CRM / distribution only