Skip to slide
18 / 31

CH 03 · The Taxonomy of Agentic RAG Systems · 07/10

Read as one page

Graph-based agentic RAG

This family combines knowledge graphs with agentic reasoning. The survey highlights Agent-G as a representative: it integrates structured graph knowledge bases with unstructured document retrieval, using modular retriever banks, a critic module, and feedback loops.

In Agent-G, graph knowledge bases supply relationships and hierarchies (for example, disease-to-symptom mappings in healthcare), unstructured documents add contextual detail, a critic module evaluates the relevance and quality of what was retrieved (flagging low-confidence results for re-retrieval), and feedback loops refine the process iteratively. The strength of this family is strong multi-hop reasoning over structured relationships combined with the flexibility of text retrieval. It suits domains where relationships are central, such as healthcare and legal analysis.

Finished this idea?

Your learning progress stays in this browser.