Knowledge Graph RAG (ToPWR)

Problem

Students and staff needed a way to quickly find information about university courses, departments, and procedures without searching through multiple systems.

Solution

Built a knowledge graph-based RAG system that converts natural language queries to Cypher queries, with guardrail-based routing to ensure safe and accurate responses.

Architecture

The system uses LangGraph for orchestrating the RAG pipeline, Neo4j as the knowledge graph store, and OpenAI GPT for response generation. MCP (Model Context Protocol) handles the tool calling.

Tech Stack

Key Challenges

• Designing efficient graph schemas for university data
• Handling ambiguous queries with multiple interpretations
• Ensuring response accuracy with guardrails