SEGRA: Structured Experience-Guided Graph Reasoning Agent for Gremlin Based Question Answering

SEGRA utilizes a curriculum-bootstrapped skill library and execution-aware verification to convert natural language into Gremlin queries for enterprise IT knowledge graphs. On IT support benchmarks, the agent achieves a 7.0x higher mean judge score than backbone chain-of-thought prompting. Implementing the skill library reduces LLM calls by 20%.