HKNQTWZcfilorux0369γζADG
QTWZcfilorux0369γζ
JMPSVYbehknqtwz258βεθCFILORUXa
HKNQTWZcfilo
cfilorux0369γ
GJMPSVYbehknqtw
SVYbehknqtwz258βεθCFILORUXadg
Healthcare AI · Polypharmacy · Clinical Reasoning

ARIA

Adaptive Risk Intelligence for Polypharmacy Assessment

An AI agent system that does not just detect drug interactions. It reasons about them.

🚨

The Problem Nobody Has Solved

A $42 billion global crisis

$42B
annual cost
59%
of frail elderly
90%+
alerts overridden

Medication errors cost the world an estimated $42 billion USD every year. That is nearly 1% of total global health expenditure. Half of all preventable harm in medical care is medication-related.

The driver is polypharmacy: the simultaneous use of five or more medications. Global prevalence sits at 37% in the general population. It climbs to 52% among inpatients and 59% among frail elderly patients.

Existing tools are static lookup tables. They emit identical, context-free warnings for every patient. The result is predictable: clinicians override more than 90% of all drug interaction alerts. The tools designed to protect patients have become background noise.

What ARIA Does Differently

Not a drug interaction checker. A clinical reasoning engine.

🕸️

N-Drug Interaction Graph

Full interaction graph with hub drug identification. Detects emergent 3-drug interactions invisible to pairwise checkers.

⏱️

Temporal Cascade Modeling

Models when an interaction peaks, not just whether it exists. Predicts critical intervention windows.

👤

Patient Phenotype Risk

Personalized risk score adjusted for age, CKD stage, hepatic function, sex, weight, and smoking status.

💊

Deprescribing Optimizer

Prioritized, actionable plan: which drug to address first, what to substitute, expected risk reduction at each step.

8 Capabilities That Don't Exist Anywhere Else

Temporal Cascade Modeling

Predicts when risk peaks over time

Pharmacokinetic Reasoning

CYP enzyme, renal, microbiome pathways via Gemini 2.5 Pro

Patient Phenotype Multiplier

Same drugs, different patients, different risk scores

N-Drug Graph + Hub ID

Beyond pairwise. Finds the one drug causing 60% of conflicts.

Evidence Grading

A–D grades with confidence scores and PubMed citations

Cumulative Burden Scores

Anticholinergic, sedation, QT. Validated clinical metrics.

Deprescribing Optimizer

Not a warning. A plan with substitutes and timelines.

Integrated Pipeline

All 8 capabilities orchestrated into one coherent report

How It Works

Three steps to a clinical reasoning report

01

Enter Medications

Input the patient's medication list and clinical context: age, CKD stage, hepatic function, comorbidities.

02

AI Reasoning Pipeline

ARIA's agent orchestrates 8 tools: RxNorm normalization, interaction detection, phenotype scoring, temporal modeling, evidence grading, and more.

03

Actionable Report

Receive a 3D interactive clinical report with risk scores, deprescribing plans, evidence citations, and exportable PDF/HTML reports.

Technology Stack

MCP Server
Rust
A2A Agent
Python + LangGraph
LLM
Gemini 2.5 Pro
Frontend
Next.js 14 + R3F
3D
React Three Fiber
Deploy
Cloud Run + Vercel

Built By

One engineer. Three weeks. Eight clinical capabilities.

Solo Builder · Hackathon Submission
W

Wiqi Lee

Data Scientist · AI/ML Researcher · Software Engineer

“An agent that thinks like a clinical pharmacologist, not a database lookup.”

Submitted to the Agents Assemble: Healthcare AI Endgame Hackathon

Sponsored by Prompt Opinion (Darena Health)