Knowledge Graph RAG with Verifiable Citations

A Streamlit application demonstrating how Knowledge Graph-based Retrieval-Augmented Generation (RAG) provides multi-hop reasoning with fully verifiable source attribution.

🎯 What Makes This Different?

Traditional vector-based RAG finds similar text chunks, but struggles with:

• Questions requiring information from multiple documents
• Complex reasoning chains
• Providing verifiable sources for each claim

Knowledge Graph RAG solves these by:

1. Building a structured graph of entities and relationships from documents
2. Traversing connections to find related information (multi-hop reasoning)
3. Tracking provenance so every claim links back to its source

✨ Features

🚀 Quick Start

Prerequisites

1. Ollama - Local LLM inference

# Install from https://ollama.ai
ollama pull llama3.2

2. Neo4j - Knowledge graph database

# Using Docker
docker run -d \
  --name neo4j \
  -p 7474:7474 -p 7687:7687 \
  -e NEO4J_AUTH=neo4j/password \
  neo4j:latest

Installation

# Clone and navigate
cd knowledge_graph_rag_citations

# Install dependencies
pip install -r requirements.txt

# Run the app
streamlit run knowledge_graph_rag.py

📖 How It Works

Step 1: Document → Knowledge Graph

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│   Document      │ ──► │  LLM Extraction  │ ──► │ Knowledge Graph │
│   (Text/PDF)    │     │  (Entities+Rels) │     │    (Neo4j)      │
└─────────────────┘     └──────────────────┘     └─────────────────┘

The LLM extracts:

• Entities: People, organizations, concepts, technologies
• Relationships: How entities connect (e.g., "works_for", "created", "uses")
• Provenance: Source document and chunk for each extraction

Step 2: Query → Multi-hop Traversal

┌─────────┐     ┌─────────────┐     ┌─────────────┐     ┌───────────┐
│  Query  │ ──► │  Find Start │ ──► │  Traverse   │ ──► │  Context  │
│         │     │   Entities  │     │  Relations  │     │  + Sources│
└─────────┘     └─────────────┘     └─────────────┘     └───────────┘

Step 3: Answer → Verified Citations

┌─────────────┐     ┌─────────────┐     ┌──────────────────┐
│   Context   │ ──► │  Generate   │ ──► │  Answer with     │
│ + Sources   │     │   Answer    │     │  [1][2] Citations│
└─────────────┘     └─────────────┘     └──────────────────┘
                                                │
                                                ▼
                                        ┌──────────────────┐
                                        │ Citation Details │
                                        │ • Source Doc     │
                                        │ • Source Text    │
                                        │ • Reasoning Path │
                                        └──────────────────┘

🖥️ Usage Example

1. Add a Document

Paste or select a sample document. The system extracts entities and relationships:

Document: "GraphRAG was developed by Microsoft Research. 
           Darren Edge led the project..."

Extracted:
  ├── Entity: GraphRAG (TECHNOLOGY)
  ├── Entity: Microsoft Research (ORGANIZATION)  
  ├── Entity: Darren Edge (PERSON)
  └── Relationship: Darren Edge --[WORKS_FOR]--> Microsoft Research

2. Ask a Question

Question: "Who developed GraphRAG and what organization are they from?"

3. Get Verified Answer

Answer: GraphRAG was developed by researchers at Microsoft Research [1], 
        with Darren Edge leading the project [2].

Citations:
  [1] Source: AI Research Paper
      Text: "GraphRAG is a technique developed by Microsoft Research..."
      
  [2] Source: AI Research Paper  
      Text: "...introduced by researchers including Darren Edge..."

🔧 Configuration

🏗️ Architecture

knowledge_graph_rag_citations/
├── knowledge_graph_rag.py   # Main Streamlit application
├── requirements.txt         # Python dependencies
└── README.md               # This file

Key Components

• KnowledgeGraphManager: Neo4j interface for graph operations
• extract_entities_with_llm(): LLM-based entity/relationship extraction
• generate_answer_with_citations(): Multi-hop RAG with provenance tracking

🎓 Learn More

This example is inspired by VeritasGraph, an enterprise-grade framework for:

• On-premise knowledge graph RAG
• Visual reasoning traces (Veritas-Scope)
• LoRA-tuned LLM integration

📝 License

MIT License