RAGU: A Multi-Step GraphRAG Engine Powered by a Compact Domain-Adapted LLM

RAGU is an open-source modular GraphRAG engine that separates knowledge-graph extraction from consolidation. Its pipeline uses two-stage typed extraction, DBSCAN-based deduplication, LLM summarization, and Leiden community detection. The authors introduce Meno-Lite-0.1, a 7B model optimized for language skills rather than broad factual knowledge, reporting a 12.5% relative harmonic-mean improvement over Qwen2.5-32B for knowledge-graph construction and parity on English GraphRAG tasks. On GraphRAG-Bench Medical, RAGU reportedly achieves evidence recall up to 0.84, runs on a single GPU, and is released under the MIT license.