Knowledge-Graph-from-Text Builder, Evaluation

Overview

A clone of Neo4j Labs’ open-source LLM Graph Builder, which turns unstructured documents into a Neo4j knowledge graph using LLM extraction. Retained for evaluation; the repository carries extensive upstream history with no studio-authored changes.

Why It Exists

To evaluate LLM-driven knowledge-graph construction and graph-RAG, extracting entities and relationships from documents into Neo4j and querying them, as a stronger alternative to flat vector RAG for connected data.

What We Built

No custom development; an upstream snapshot used as a working reference. Observed architecture: a FastAPI backend that orchestrates LLM extraction into Neo4j, a React frontend for ingestion/visualization, LangChain integration, and Docker Compose deployment.

Technologies & Approach

Python/FastAPI + React on a Neo4j graph store, using LLMs to extract entities/relationships and build a queryable knowledge graph for retrieval-augmented generation.

Outcome / Impact

Gave the studio a hands-on reference for graph-RAG and unstructured-to-graph pipelines, informing related internal work (e.g. GraphRAG over technical documents). Documented as evaluation/R&D.

Capabilities Demonstrated

• Evaluating unstructured-text → knowledge-graph pipelines
• Understanding graph-RAG architectures on Neo4j