RxGuard

RxGuard is a policy-enforced, evidence-linked medication safety layer that cross-references user context against official FDA drug labels. It is designed to intercept and correct unsafe AI guidance using deterministic rules and verifiable citations.

Summary

• Deterministic policy decisions: BLOCK, CAUTION, CLARIFY, INFO
• FDA label pinning with openFDA snapshots
• Verifiable evidence quotes (substring-verified)
• Profile-aware risk checks (pregnancy, conditions, current medications)
• Optional side-by-side baseline model outputs for demo comparison

Quick Start

Prerequisites

• Node.js 18+
• Firebase project with Storage enabled
• (Optional) openFDA API key
• (Optional) Local LLM server for demo outputs

Install

git clone https://github.com/yourusername/rxguard.git
cd rxguard/RxGuard_Wellthorav2
npm install
npm i firebase-admin

Configure

Copy the environment template and set values:

cp rxguard.env.example .env.local

Example .env.local:

FIREBASE_STORAGE_BUCKET=your-project.appspot.com
GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json

OPENFDA_API_KEY=your_api_key

RXGUARD_LLM_PROVIDER=openai_compat
RXGUARD_LLM_BASE_URL=http://localhost:8000
RXGUARD_MISTRAL_MODEL=mistral-7b
RXGUARD_LLAMA_MODEL=llama3-8b

Run

npm run dev

Open http://localhost:3000/dashboard/rxguard.

Architecture

1. Resolve drug name to a pinned FDA label snapshot
2. Evaluate deterministic policy rules against label evidence and user profile
3. Generate a proof card with verified evidence quotes
4. Optionally request baseline model outputs for side-by-side comparison

Decision Types

Decision	Risk	Meaning
BLOCK	HIGH	Strong contraindication detected
CAUTION	MODERATE	Potential risks identified
CLARIFY	UNKNOWN	Missing or ambiguous inputs
INFO	LOW	No high-severity matches found

Project Structure

RxGuard_Wellthorav2/
	src/
		ai/flows/                 # Genkit flow integration
		app/dashboard/rxguard/    # UI route
		lib/rxguard/              # Core engine (policy, evidence, FDA)
	scripts/rxguard/            # Utilities (seedSyntheticProfiles)
	rxguard.env.example
	RXGUARD_INSTALL.md

API Usage

import { rxguardAnswer } from '@/lib/rxguard/rxguard';

const response = await rxguardAnswer({
	question: 'Is Advil safe for my headache?',
	primaryDrug: 'Advil',
	otherMeds: ['Warfarin'],
	profile: {
		pregnancy: 'unknown',
		conditions: ['stomach ulcers'],
		currentMeds: ['Warfarin'],
	},
	includeNakedModelAnswer: false,
});

RxBench Dataset

The repository includes a dataset of FDA-label-derived medication safety scenarios:

RxBench 200k FDA-Label–Derived Medication Safety Scenarios.jsonl

Use it to benchmark policy coverage and evaluate guardrail performance.

Disclaimers

RxGuard is for educational and research use only. It does not provide medical advice and cannot confirm medication safety. Always consult a licensed clinician or pharmacist for medication decisions.

Development Notes

Seed synthetic profiles:

npx tsx scripts/rxguard/seedSyntheticProfiles.ts

Genkit dev (optional): ensure src/ai/dev.ts imports @/ai/flows/rxguard-answer-flow.ts.

License

MIT License. See the LICENSE file for details.