Multi-Agent LLM System for Academic Graph Exploration

Inspiration

I've always enjoyed graph analytics and recently got into the world of LLMs. It was exciting to see their combination in this hackathon!

What it does

This notebook demonstrates a multi-agent workflow for analyzing academic networks, integrating:

• ArangoDB for graph storage and querying with AQL and SearchView,
• NetworkX with CUDA for efficient large-scale network analysis,
• LangChain for converting user queries into structured AQL and NetworkX operations,
• LangGraph for orchestrating agents that handle query decomposition, data retrieval, analysis, and visualization.

How we built it

I started by exploring a fantastic large-scale open dataset with millions of nodes and edges related to academic networks. From there, I iteratively built and tested LangGraph-based functions to process queries and analyze the data.

Challenges we ran into

Understanding the connections and dependencies between LangChain and LangGraph wasn't always intuitive, requiring several iterations to get the workflow right.

Accomplishments that we're proud of

I successfully implemented dynamic query routing with LangGraph and automatically labelling communities in the network with LLM.

What we learned

I gained hands-on experience using ArangoDB for querying large-scale graphs and LangGraph for orchestrating multi-agent workflows.

What's next for Multi-Agent LLM System for Academic Graph Exploration

I plan to test and integrate additional NetworkX functions to expand the system’s analytical capabilities.

Built With

• arangodb
• langchain
• langgraph
• networkx
• python