Next-Gen Indonesia IT Law Q&A: Knowledge Graphs & LLMs

Information Systems Department, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia.

Inspiration

Understanding Indonesia's IT laws can be complex due to the vast number of regulations, articles, and interconnections. By leveraging Retrieval-Augmented Generation (RAG), Large Language Models (LLMs), and Knowledge Graphs, we aim to provide a smarter, context-aware Q&A system that enables legal professionals, researchers, and policymakers to retrieve precise legal insights efficiently.

What it does

This system allows users to:

• Perform AQL (ArangoDB Query Language) and semantic searches across 63 Indonesia IT law regulations.
• Identify the most influential regulations and explore their relationships.
• Analyze connections between regulations and articles using graph-based analytics.
• Visualize legal structures with matplotlib for better understanding.

Installation

To install the latest version of this code, please use this following command.

1. Git clone this project to the local computer

   git clone https://github.com/bayu-siddhi/graph-rag-arangodb
   cd graph-rag-arangodb

2. Create a virtual environment (optional but recommended):.

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

3. Install all the project dependencies.

   pip install -r requirements.txt

4. Copy the environment configuration.

   cp .env.example .env

5. Run the Gradio app on the main.py file.

   python main.py

   Or see the development process in notebook.ipynb.

Note

• This project was developed using python==3.11.4, see requirements.txt for dependencies details.
• Edit the .env file and fill in your ArangoDB credentials and other necessary configuration.
• Make sure ArangoDB is running and that the user has Administrate privileges.

How we built it

1. Knowledge Graph Construction:
   • Built using ArangoDB with 63 regulations, 2,423 articles, and over 7,500 relationships including amendments, references, and hierarchical structures.
2. Graph Analytics:
   • Integrated NetworkX to analyze shortest paths, centrality, and community structures.
3. Natural Language Processing:
   • Utilized RAG with LLMs to provide context-aware responses based on graph retrieval.
4. User Interface:
   • Developed with Gradio, running locally at http://127.0.0.1:7861.

Challenges we ran into

• Efficiently indexing and querying large-scale legal documents while maintaining response speed.
• Integrating graph-based retrieval with LLM-generated responses.
• Designing an intuitive legal Q&A interface for non-technical users.

Accomplishments that we're proud of

• Successfully integrated ArangoDB, NetworkX, and LLMs for graph-enhanced legal Q&A.
• Enabled advanced legal reasoning by mapping the impact and references between laws.
• Built an interactive visualization that helps users see legal connections dynamically.

What we learned

• The power of RAG and LLMs in legal document retrieval.
• How knowledge graphs enhance Q&A accuracy by structuring unstructured data.
• The importance of graph analytics in legal research, especially in identifying key regulatory influences.

What's next for Next-Gen Indonesia IT Law Q&A: Knowledge Graphs & LLMs

• Expanding the dataset to include more legal domains beyond IT law.
• Enhancing real-time graph analytics for deeper legal insights.
• Deploying the system as a cloud-based API for broader accessibility.
• Improving natural language understanding to support multi-turn legal conversations.