RxAccess AI — Intelligent Prescription Access & Adherence Platform

🚀 Try it live: https://rxaccess-ai.streamlit.app

---

🎯 Project Overview

RxAccess AI is a production-grade healthcare AI prototype that demonstrates end-to-end capabilities for prescription management, patient access, prior authorization, affordability insights, and adherence intelligence. This platform closely mirrors real-world challenges solved by health-tech companies like PHIL.

🏥 Alignment with PHIL's Mission

This platform addresses key healthcare challenges:

• Prescription Access: Intelligent OCR and extraction from images/PDFs
• Prior Authorization: Automated PA form generation, status tracking, and approval prediction
• Affordability: Insurance coverage estimation and patient assistance program recommendations
• Adherence Intelligence: ML-powered risk prediction and personalized interventions
• Multi-Stakeholder Support: Dashboards for patients, providers, pharmacies, and pharma companies

🏗️ Architecture

graph TB
    A[User Upload] --> B[OCR Engine]
    B --> C[LLM Extraction & Structuring]
    C --> D[Structured Prescription Data]
    D --> E[RAG Medical Assistant]
    D --> F[PA Assistant]
    D --> G[Affordability Engine]
    D --> H[Adherence Predictor]
    
    E --> I[ChromaDB Vector Store]
    F --> J[PA Form Generator]
    G --> K[Coverage Estimator]
    H --> L[ML Model XGBoost]
    
    I --> M[Patient Dashboard]
    J --> M
    K --> M
    L --> M
    
    M --> N[Provider View]
    M --> O[Pharma Insights]
    
    style A fill:#e1f5ff
    style M fill:#fff4e1
    style L fill:#ffe1f5

Loading

🚀 Features

1. Prescription Ingestion & Extraction

• Upload prescription images (JPG, PNG) or PDFs
• OCR using Tesseract with LLM-powered correction
• Extract: medicine name, dosage, frequency, duration, doctor info, patient details
• Structured JSON output with confidence scores

2. RAG-Powered Medical Assistant

• Ask questions about uploaded prescriptions
• Knowledge base with drug information, interactions, side effects
• ChromaDB vector store for semantic search
• Context-aware responses using LangChain

3. Prior Authorization Assistant

• Automated PA form generation
• Required documentation checklist
• Approval likelihood prediction
• Status tracking (Pending → Under Review → Approved/Denied)
• Missing information alerts

4. Affordability & Access Intelligence

• Insurance coverage estimation
• Copay calculator
• Patient assistance program recommendations
• Cash-pay vs insurance comparison
• Generic alternatives suggestions

5. Adherence Prediction & Personalization

• XGBoost ML model for adherence risk scoring
• Features: age, medication class, regimen complexity, past adherence
• Personalized intervention generation
• Reminder scheduling with motivational messaging

6. Multi-Stakeholder Dashboards

• Patient View: Prescription details, Q&A, adherence score, reminders
• Provider/Pharmacy View: Extracted data, PA status, patient insights
• Pharma Insights: Aggregated metrics, adherence rates, PA success rates

🛠️ Tech Stack

Component	Technology
Backend	Python 3.10+, FastAPI
Frontend	Streamlit
LLM Framework	LangChain
LLM Provider	Ollama (local) / Groq / OpenAI
Vector Store	ChromaDB
OCR	Tesseract + LLM correction
ML	scikit-learn, XGBoost
Deployment	Docker, AWS-ready

📦 Installation

Prerequisites

• Python 3.10+
• Docker (optional)
• Tesseract OCR

Local Setup

1. Clone the repository

git clone <repository-url>
cd rxaccess-ai

2. Install Tesseract OCR

Windows:

# Download from: https://github.com/UB-Mannheim/tesseract/wiki
# Add to PATH

macOS:

brew install tesseract

Linux:

sudo apt-get install tesseract-ocr

3. Create virtual environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

4. Install dependencies

pip install -r requirements.txt

5. Set up environment variables

cp .env.example .env
# Edit .env with your API keys

6. Initialize the system

python scripts/init_system.py

Docker Setup

docker build -t rxaccess-ai .
docker run -p 8501:8501 -p 8000:8000 rxaccess-ai

🎮 Usage

Start the Application

streamlit run streamlit_app/app.py

Access at: http://localhost:8501

Demo Flow

1. Upload Prescription

   • Navigate to "📄 Upload Prescription" tab
   • Upload image or PDF
   • View extracted structured data

2. Ask Questions

   • Go to "💬 Medical Assistant" tab
   • Ask about side effects, interactions, dosage instructions
   • Get AI-powered responses with sources

3. Check Prior Authorization

   • Visit "📋 Prior Authorization" tab
   • Review PA form summary
   • Check approval likelihood
   • Track submission status

4. Explore Affordability

   • Open "💰 Affordability" tab
   • View insurance coverage estimate
   • Compare cash-pay options
   • Find patient assistance programs

5. View Adherence Insights

   • Check "📊 Adherence Intelligence" tab
   • See risk score prediction
   • Review personalized interventions
   • Set up reminders

6. Multi-Stakeholder Views

   • Switch between Patient, Provider, and Pharma dashboards
   • Explore role-specific insights

📁 Project Structure

rxaccess-ai/
├── streamlit_app/
│   ├── app.py                      # Main Streamlit application
│   ├── pages/
│   │   ├── 1_upload.py            # Prescription upload
│   │   ├── 2_assistant.py         # RAG medical assistant
│   │   ├── 3_prior_auth.py        # PA assistant
│   │   ├── 4_affordability.py     # Affordability engine
│   │   ├── 5_adherence.py         # Adherence intelligence
│   │   └── 6_dashboards.py        # Multi-stakeholder views
│   └── components/
│       ├── sidebar.py             # Shared sidebar
│       └── utils.py               # UI utilities
├── src/
│   ├── extraction/
│   │   ├── ocr_engine.py          # Tesseract OCR
│   │   └── llm_extractor.py       # LLM-based extraction
│   ├── rag/
│   │   ├── vector_store.py        # ChromaDB setup
│   │   ├── retriever.py           # RAG retriever
│   │   └── qa_chain.py            # Q&A chain
│   ├── prior_auth/
│   │   ├── pa_generator.py        # PA form generation
│   │   ├── approval_predictor.py  # Approval likelihood
│   │   └── status_tracker.py      # Status management
│   ├── affordability/
│   │   ├── coverage_estimator.py  # Insurance coverage
│   │   └── assistance_finder.py   # Patient assistance
│   ├── adherence/
│   │   ├── risk_predictor.py      # ML risk model
│   │   ├── intervention_gen.py    # Personalized interventions
│   │   └── model_trainer.py       # Model training
│   ├── utils/
│   │   ├── pii_redaction.py       # PII handling
│   │   ├── disclaimer.py          # Legal disclaimers
│   │   └── logger.py              # Logging setup
│   └── config.py                  # Configuration
├── backend/
│   ├── main.py                    # FastAPI application
│   ├── routes/
│   │   ├── extraction.py          # Extraction endpoints
│   │   ├── rag.py                 # RAG endpoints
│   │   ├── prior_auth.py          # PA endpoints
│   │   └── adherence.py           # Adherence endpoints
│   └── models/
│       └── schemas.py             # Pydantic models
├── models/
│   ├── adherence_model.pkl        # Trained XGBoost model
│   └── scaler.pkl                 # Feature scaler
├── data/
│   ├── knowledge_base/
│   │   ├── drug_info.json         # Drug information
│   │   ├── interactions.json      # Drug interactions
│   │   └── side_effects.json      # Side effects database
│   ├── synthetic/
│   │   ├── prescriptions/         # Sample prescriptions
│   │   ├── patient_data.csv       # Synthetic patient data
│   │   └── adherence_data.csv     # Training data
│   └── uploads/                   # User uploads
├── evaluation/
│   ├── extraction_eval.py         # OCR accuracy metrics
│   ├── rag_eval.py                # RAG faithfulness
│   ├── model_eval.py              # ML model performance
│   └── results/                   # Evaluation results
├── scripts/
│   ├── init_system.py             # System initialization
│   ├── generate_synthetic_data.py # Data generation
│   └── train_adherence_model.py   # Model training
├── docs/
│   ├── ARCHITECTURE.md            # Detailed architecture
│   ├── API.md                     # API documentation
│   ├── DEPLOYMENT.md              # Deployment guide
│   └── SECURITY.md                # Security considerations
├── tests/
│   ├── test_extraction.py
│   ├── test_rag.py
│   ├── test_prior_auth.py
│   └── test_adherence.py
├── .env.example                   # Environment template
├── .gitignore
├── requirements.txt
├── Dockerfile
├── docker-compose.yml
└── README.md

🔧 Configuration

Edit .env file:

# LLM Configuration
LLM_PROVIDER=ollama  # ollama, groq, openai
OLLAMA_MODEL=llama3.1
GROQ_API_KEY=your_groq_key
OPENAI_API_KEY=your_openai_key

# Vector Store
CHROMA_PERSIST_DIR=./data/chroma_db

# OCR
TESSERACT_PATH=/usr/bin/tesseract

# AWS (Optional)
AWS_ACCESS_KEY_ID=your_key
AWS_SECRET_ACCESS_KEY=your_secret
AWS_REGION=us-east-1

# Application
DEBUG=True
LOG_LEVEL=INFO

📊 Evaluation Metrics

Extraction Accuracy

• Character Error Rate (CER)
• Word Error Rate (WER)
• Field-level accuracy (medicine name, dosage, etc.)

RAG Performance

• Faithfulness score
• Answer relevancy
• Context precision/recall

ML Model Performance

• Adherence prediction: AUC-ROC, F1-score, precision, recall
• Feature importance analysis

Run evaluation:

python evaluation/run_all_evals.py

🔒 Security & Compliance

HIPAA Considerations

• PII redaction for sensitive data
• Encrypted data storage (implement in production)
• Audit logging for all access
• Role-based access control
• Secure API endpoints with authentication

Disclaimers

• "Not medical advice" disclaimer on all outputs
• "For demonstration purposes only" notice
• Recommendation to consult healthcare professionals

🚀 Deployment

AWS Deployment

1. S3 for file storage

aws s3 mb s3://rxaccess-ai-uploads

2. ECR for Docker images

aws ecr create-repository --repository-name rxaccess-ai
docker tag rxaccess-ai:latest <account>.dkr.ecr.us-east-1.amazonaws.com/rxaccess-ai
docker push <account>.dkr.ecr.us-east-1.amazonaws.com/rxaccess-ai

3. ECS/Fargate for containers

# Use provided CloudFormation template
aws cloudformation create-stack --stack-name rxaccess-ai --template-body file://deploy/cloudformation.yml

See docs/DEPLOYMENT.md for detailed instructions.

🧪 Testing

# Run all tests
pytest tests/

# Run specific test suite
pytest tests/test_extraction.py -v

# Run with coverage
pytest --cov=src tests/

📈 Future Improvements

Short-term

• Real-time prescription verification with pharmacy databases
• Multi-language support for prescriptions
• Mobile app integration
• SMS/Email reminder system
• Integration with EHR systems (HL7 FHIR)

Medium-term

• Advanced PA automation with payer API integration
• Real-time insurance eligibility verification
• Predictive analytics for medication shortages
• Blockchain for prescription authenticity
• Telemedicine integration

Long-term

• Clinical trial matching based on prescriptions
• Pharmacogenomics integration
• Real-world evidence generation
• Population health analytics
• AI-powered formulary optimization

🤝 Contributing

Contributions are welcome! Please read our contributing guidelines and submit pull requests.

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

• Inspired by PHIL's mission to improve prescription access and adherence
• Built with open-source AI/ML tools
• Synthetic data generated for demonstration purposes

📞 Contact

For questions or support, please open an issue or contact the development team.

---

⚠️ IMPORTANT DISCLAIMER: This is a prototype for demonstration purposes only. Not intended for actual medical use. Always consult healthcare professionals for medical advice.