🔍 CodinzHub GraphRAG - Knowledge Graph Retrieval-Augmented Generation (RAG) with Verifiable Citations

Developed by Coder Coder | 🚀 Find more projects at CodinzHub Projects

An advanced Streamlit engine demonstrating how Knowledge Graph-based Retrieval-Augmented Generation (RAG) resolves multi-hop reasoning challenges with source provenance tracking.

🛠️ Technology Stack

UI Layer: Streamlit (Python)
Graph Storage: Neo4j Graph Database
Cognitive Inference: Ollama (Running Mistral or Llama)
API Integration: Python Neo4j Driver & Ollama Python SDK

🎯 Architectural Overview & Core Value

Standard semantic vector search matches text chunks based on keyword/semantic similarity, but fails at:

Connecting disparate facts across multiple documents (multi-hop traversal).
Outlining reasoning traces showing exactly how a decision was formulated.
Attriburing verifiable source claims to the generated output.

CodinzHub GraphRAG addresses these challenges by:

Parsing Entities & Relationships: Building structured GraphNodes and GraphEdges from raw input files.
Deep Path Traversal: Traversing relational links inside Neo4j to find indirect associations.
Provenance Validation: Mapping inline answers directly to original snippets.

🚀 Quick Setup & Ingestion

1. Start Local Prerequisites

Ollama Inference Engine:
```
ollama pull llama3.2
```

Neo4j Graph Database Instance (Docker):

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

2. Configure Environment & Script

Navigate to the folder:

cd codinzhub-skills/codinzhub-rag/codinzhub-graph-rag

Load the dependencies:
```
pip install -r requirements.txt
```
Launch the local Streamlit panel:
```
streamlit run codinzhub-graph-rag.py
```

📖 Operational Pipeline

Ingestion Pipeline

flowchart TD
    Doc[Raw Text/PDF Document] -->|Entity Parsing| LLM(Local Ollama Model)
    LLM -->|Extract JSON Nodes & Edges| Parser[Structure Parser]
    Parser -->|Add GraphNode| DB[(Neo4j Database)]
    Parser -->|Add GraphEdge| DB

Retrieval & Attribution Flow

flowchart LR
    Query[User Question] -->|Text Index Match| Search[Semantic Search]
    Search -->|Multi-Hop Path Traversal| Expand[Depth-2 Connections]
    Expand -->|Formatted Context Mapping| Gen[Attributed Generator]
    Gen -->|Provenace Attributions| Out[Answer + Citations]

🖥️ Production Ingestion Example

Step 1: Input Document

CloudEngine is an automated deployment framework designed by CodinzHub Labs. It uses 
generative configuration templates to provision server clusters. The system was 
architected by Coder Coder, who led the development team in San Francisco.

Step 2: Extraction Output

Entity Node: CloudEngine (TECHNOLOGY)
Entity Node: CodinzHub Labs (ORGANIZATION)
Entity Node: Coder Coder (PERSON)
Relationship Link: Coder Coder --[ARCHITECTED]--> CloudEngine
Relationship Link: CloudEngine --[DESIGNED_BY]--> CodinzHub Labs

⚙️ Configuration Properties

Neo4j Default Bolt URI: bolt://127.0.0.1:7687
Neo4j Default User: neo4j
LLM Local Model: llama3.2

CodinzHub GraphRAG