Answer: Snowflake described approaches for integrating formal ontologies into Snowflake to ground Cortex Agents in domain concepts and relationships, and it compared ontology-aware approaches against a baseline semantic-view agent using a simplified biomedical dataset. Snowflake said it evaluated knowledge graphs, GraphRAG, and targeted terminology mappings, and reported benchmark results that it presented as a blueprint for building more precise and explainable enterprise agents.
Snowflake said the comparison included a baseline semantic-view agent and ontology-aware variants that use knowledge graphs, GraphRAG, and targeted terminology mappings to represent domain meaning beyond schemas. Snowflake framed the work around grounding agents in domain concepts and relationships in order to support reasoning in the evaluated setup.
Source: Snowflake