GraphRAG-based AI Research Assistant

A multi-agent system for automated literature review using LangGraph and Neo4j knowledge graphs.

This project presents a multi-agent system designed to automate the process of academic literature review through structured retrieval and reasoning pipelines.

System Architecture

The system is built on LangGraph, enabling orchestration of multiple specialised agents for planning, retrieval, and synthesis tasks. The pipeline follows a multi-step reasoning workflow:

  1. Planning Agent — decomposes a user research query into structured sub-tasks.
  2. Retrieval Agent — performs hybrid search (dense + sparse) over a corpus of academic documents.
  3. GraphRAG Module — traverses a knowledge graph in Neo4j to surface related concepts, entities, and cross-paper relationships not captured by vector similarity alone.
  4. Synthesis Agent — consolidates retrieved evidence into a coherent, cited research summary.

Key Contributions

  • Designed and implemented the full multi-agent orchestration pipeline using LangGraph.
  • Constructed the GraphRAG module combining hybrid semantic search with knowledge graph traversal.
  • Integrated Neo4j as the graph backend, enabling entity-relationship queries across the document corpus.

Technologies

Python · LangGraph · Neo4j · Large Language Models (LLMs)