Vanilla RAG retrieves chunks.
GraphRAG follows the chain — and it shows.

Same corpus, same questions, same local generator — the only variable is retrieval. Vanilla gets top-k chunks by cosine; the hybrid branch adds Neo4j traversal ≤3 hops from entities found in the question. Accuracy is deterministic: exact gold-string containment, no LLM judge, misses published.

Full disclosure: the corpus is a 28-document synthetic company (Auralis Systems GmbH — fictional, written for this demo) with multi-hop fact chains deliberately split across documents. The knowledge graph runs in a local Neo4j (Docker) — it is not hosted; this page replays recorded runs. Entity extraction: Gemini (model cascade); embeddings and answers: local Ollama models.
Vanilla RAG top-4 cosine
overall accuracy
single-hop
multi-hop
Hybrid GraphRAG top-3 + graph facts
overall accuracy
single-hop
multi-hop

Every question, both branches

20 gold questions — 10 single-hop, 10 multi-hop. Click one to see both answers, the docs each branch retrieved, and the graph path the hybrid branch walked.

Where the graph won

Multi-hop questions where vanilla missed and the hybrid branch answered — with the traversal chain that made the difference.

Methodology & honest limitations