VDGR-RAG: Vectors, Directories, Graphs, and Reflection Are All You Need for Unified Reasoning over Hierarchical Enterprise Knowledge

Retrieval-Augmented Generation (RAG) is essential for enterprise knowledge question answering (QA), particularly in domains with complex product documentation like telecommunications. However, existing approaches overlook the holistic integration of diverse retrieval strengths, leading to inaccurate domain routing, poor utilization of hierarchical structures, and limited reasoning capabilities. We present VDGR-RAG, an agentic GraphRAG system integrating vector retrieval, directory-driven reasoning, graph traversal, and iterative reflection. VDGR-RAG constructs a Hierarchical Heterogeneous Knowledge Graph ($\text{H}^2\text{KG}$) preserving both directory structures and semantic relationships, then employs four composable tools for retrieval: (1) directory-enhanced routing, (2) multi-route retrieval combining vector-, table-of-contents (TOC)-, and graph-based search, (3) directory backtracking, and (4) dynamic reflection. Experiments on enterprise product documents across four wireless domains demonstrate that VDGR-RAG significantly outperforms various RAG baselines in both retrieval recall and QA accuracy.