Clinical Intelligence RAG 🏥 🤖

A privacy-first RAG pipeline designed for healthcare, implementing PII scrubbing and AWS Bedrock integration to ensure HIPAA-compliant AI inference.

Quick Start

docker-compose up --build
# API:  http://localhost:8000/docs
# UI:   http://localhost:8501

Project Architecture

graph TD
    A["📄 Clinical PDFs"] --> B["🛠️ Ingestion<br/>(Parse & Chunk)"]
    B -->|PHI Scrubbing| C["🧬 Embeddings<br/>(Multi-Provider)"]
    C --> D[("🌲 Pinecone<br/>Vector DB")]
    
    subgraph RAG["⚙️ RAG Engine"]
    D <-->|Semantic Search| E["🔍 Retrieval<br/>& Compression"]
    E -->|Relevant Context| F["🏭 LLM Provider<br/>Selection"]
    F -->|OpenAI/Anthropic/Bedrock| G["🧠 Generate<br/>Answer"]
    end
    
    G -->|Grounded Response| H["🚀 FastAPI<br/>REST API"]
    
    subgraph UI["👥 User Interface Layer"]
    H --> I["💻 Streamlit<br/>Dashboard"]
    I --> J["🏥 Clinical<br/>Intelligence"]
    end
    
    style A fill:#e8f5e9
    style D fill:#1b5e20,stroke:#fff,color:#fff,stroke-width:3px
    style G fill:#4a148c,stroke:#fff,color:#fff,stroke-width:3px
    style RAG fill:#f5f5f5,stroke:#666,stroke-width:2px
    style UI fill:#fff3e0,stroke:#ff6f00,stroke-width:2px
    style H fill:#ff9800,stroke:#fff,color:#fff,stroke-width:2px
    style I fill:#ffb74d,stroke:#fff,color:#333,stroke-width:2px
    style J fill:#0d2949,stroke:#fff,color:#fff,stroke-width:3px

Loading

Features

• 📄 Upload PDFs
• 💬 Ask questions → AI answers from docs
• 📥 Export analysis
• 🔐 De-identify PHI (HIPAA)
• ⚡ 20-30% fewer tokens (compression)
• 🔄 Any LLM (swap in .env)
• ✅ Validated (F:1.00, P:1.00)
• 📊 Production Monitoring** (LangSmith tracing)

Quality

Metric	Score
Faithfulness	1.00
Relevancy	0.97
Precision	1.00
Overall	0.99

Docs

• SETUP.md - Install
• API.md - REST API
• UI.md - Dashboard
• FEATURES.md - Advanced
• MONITORING.md - 📊 LangSmith Tracing
• TROUBLESHOOTING.md - Help
• DOCKER.md - Deploy
• ARCHITECTURE.md - Design

Tech Stack

Layer	Technology	Purpose
Backend API	FastAPI	REST endpoints, async request handling
Frontend UI	Streamlit	Interactive dashboard, document upload
LLM Orchestration	LangChain	Chain-of-thought reasoning, prompt management
Vector Database	Pinecone	Semantic search, embeddings storage
Evaluation	Ragas Framework	RAG quality metrics (Faithfulness, Precision, Recall)
Containerization	Docker + Compose	Production deployment, multi-service orchestration
LLM Providers	OpenAI, Anthropic, AWS Bedrock	Multi-provider support (plug-and-play)
Embeddings	OpenAI/Anthropic/Bedrock	Text-to-vector conversion
Testing	Pytest	Unit + integration tests
Language	Python 3.11+	Core implementation language
Monitoring	LangChain Tracing (LangSmith)	Debug RAG pipeline, token tracking, latency analysis

What's Inside

✅ Provider Factory (swap LLMs)
✅ Contextual Compression (20-30% tokens)
✅ PHI De-identification (HIPAA)
✅ Gold QA Benchmarks
✅ Ragas Evaluation
✅ LangSmith Monitoring** (production observability)

Help

SETUP.md • API.md • MONITORING.md • TROUBLESHOOTING.md

---

Evaluation

Run Evaluation:

python eval/evaluate_rag.py

Expected Results:

• Faithfulness: 1.00/1.00
• Answer Relevancy: 0.97/1.00
• Context Precision: 1.00/1.00
• Overall Score: 0.99/1.00

---

🩺 Troubleshooting

ValidationException (Bedrock): Ensure your AWS region supports the selected model and that you have active model access in your Bedrock console.

IndexNotFound (Pinecone): Ensure the PINECONE_INDEX_NAME in your .env matches the index you created in the Pinecone dashboard.

ModuleNotFoundError: Ensure you have activated your virtual environment:

# On Windows:
.venv\Scripts\Activate.ps1
# On Linux/Mac:
source .venv/bin/activate

No working Bedrock models found: This is expected if you don't have an AWS account or Bedrock access. The system automatically falls back to Anthropic (Claude 3.5 Sonnet) or OpenAI (GPT-4o) as configured in your .env.

---

🔍 Optional: Enable LangChain Tracing (LangSmith)

For debugging and monitoring RAG pipeline execution, you can enable LangChain tracing via LangSmith:

1. Sign up for LangSmith (free tier available)

2. Get your API key from LangSmith dashboard

3. Add to your .env:

   LANGCHAIN_TRACING_V2=true
   LANGCHAIN_API_KEY=your-langsmith-api-key

4. Run your queries - traces will automatically be sent to LangSmith:

   python test_query.py

The system will print: 🔍 LangChain tracing enabled - traces will be sent to LangSmith

Benefits:

• Monitor token usage and latency
• Debug RAG chain execution step-by-step
• Track LLM calls, embeddings, and retrieved context
• Visualize the complete prompt flow