Graph RAG with GraphDB Talk to Your Graph

Atanas Kiryakov demonstrates Graph RAG with the Talk to Your Graph (TTYG) chatbot tool of GraphDB (ver. 10.8 or later), codenamed Quadro. This is a 10-min video on how to set up Graph RAG (Type 3), where the knowledge graph is used as a content repository and a structured source of domain knowledge. The most impressive capability demonstrated is the flowless text-to-SPARQL translation happening without human intervention in a predictable and transparent manner. The test case: get NASA in the list of US government customers derived from marketing documents. Steps and challenges: 1. “Chunky” RAG doesn’t work - no mention of “government” in the chunks referring to NASA 2. Graph RAG Types 1 & 2 - there are too many chunks that mention U.S. Gov. organizations 3. Graph RAG Type 3 “Graph as a database” gets it done! The data in GraphDB: * OTKG Content Hub: 1000 Docs + Schema.org Metadata (1M statements) * Wikidata Truthy (8B statements) + materialized inference (4B statements) * Entity linking: doc, schema:mentions, entity generated with CEEL AI model See everything that GraphDB can do for you and download it here: https://hubs.la/Q02Z0Yjm0