Combining Knowledge Graphs with LLMs

Beyond flat chunks

Plain RAG retrieves isolated text chunks. A knowledge graph stores entities and their relationships, letting you answer questions that require connecting facts ("which projects use libraries that depend on X?").

GraphRAG in brief

1. Extract entities and relations from your documents (an LLM can do this).
2. Store them as a graph.
3. At query time, traverse the graph and retrieve text for grounding.

When it's worth it

Graphs add complexity, so reach for them when relationships and multi-hop reasoning matter. For simple Q&A, plain RAG is usually enough.