Retrieval-augmented generation (RAG) has become a widely adopted technique for reducing hallucinations and injecting domain knowledge into large language models. Most implementations rely on vector similarity search. This is effective for semantic matching but limited when accurate responses require understanding relationships between entities, not just proximity in embedding space.
This session introduces knowledge graphs as a complementary grounding architecture for LLM pipelines. We'll examine where vector RAG breaks down in scenarios like:
● Multi-hop reasoning
● Entity disambiguation
● Relational context
The session will demonstrate how a property graph model captures what embeddings obscure. Attendees will see how GraphRAG pipelines combine the recall strengths of vector search with the precision of graph traversal, producing responses that are more accurate and inherently more explainable.
No prior graph database experience is required, as we’ll be introducing the fundamental building blocks of the labeled property graph paradigm. Engineers familiar with RAG pipelines and transformer-based models will find the concepts immediately applicable.
Speaker(s): Rob Martin,
Virtual: https://events.vtools.ieee.org/m/557613