🧠 Multi-Modal Knowledge Graph Builder using LLMs

This project demonstrates the construction of a Knowledge Graph from multi-modal data sources like YouTube subtitles, Reddit threads, and official PDFs using LLMs and embeddings, with storage in a Neo4j graph database.

---

📦 Features

• Scrape and parse data from YouTube, Reddit, and Web PDFs
• Normalize and merge into unified JSON format
• Construct graph schema using LLMs
• Store graph data in Neo4j using Cypher
• Interface with frontend UI for chat and visualization

---

🚀 Project Structure

├── backend/         ← Python API to interface with the graph
├── common/          ← Shared utilities like LLMs and handlers
├── Data/            ← Contains scraped and merged data
├── frontend/        ← React + Tailwind frontend interface
├── graphrag/        ← Core KG builder pipeline, LLM graph builder, indexers
├── ipnyb_files/     ← Notebooks for experimentation
├── .env             ← Environment configuration
├── requirements.txt ← Python dependencies

---

🔧 Prerequisites

1. Install Neo4j and run on your local machine
   👉 Download Neo4j

2. Enable APOC Plugin (Awesome Procedures On Cypher)

   • Go to Neo4j Desktop → Plugins → Install APOC
   • Ensure it's enabled in neo4j.conf:

     dbms.security.procedures.unrestricted=apoc.*
     dbms.security.procedures.allowlist=apoc.*
     

3. Configure Azure Keys

   • Set up your Azure environment
   • Place your keys in the .env file

---

📄 .env Configuration

LLM_MODEL_NAME=deepseek-r1:7b
EMBEDDING_MODEL_NAME=nomic-embed-text
NEO4J_URI=neo4j://localhost:7687
NEO4J_USERNAME=neo4j
NEO4J_PASSWORD=
KG_BUILDER_LLM_MODEL_NAME=llama3.1:8b
KG_BUILDER_EMBEDDING_MODEL_NAME=nomic-embed-text

---

💾 Data Collection (Apple Vision Pro Example)

• YouTube: Subtitles downloaded as .srt, converted to .json
• Reddit: API bot fetches threads and comments
• Documentation: Scraper crawls and converts Apple VisionOS docs

Merged result: Data/VisionPro/merged.json

---

⚙️ How to Run

1. Populating GraphDB Using Graph-Indexing

   python graphrag/test_add_data.py

2. Creating A Vector-Index On The Populated Nodes

   python graphrag/create_vector_index.py

3. Running The CLI Application

   python3 -m streamlit run graphrag/graphrag.py

4. Running Streamlit Application

   python3 -m streamlit run app.py