Orion-RAG: Path-Aligned Hybrid Retrieval for Graphless Data

TL;DR

Orion-RAG tackles retrieval-augmented generation on graphless, fragmented data by introducing a Path-Annotation Data Augmentation system and a Multi-Layer Hybrid Retrieval framework that link disparate documents via lightweight Semantic Paths. An offline augmentation step induces a graphless structure and an online, sub-query–driven inference pipeline combines path-based retrieval, re-ranking, and pruning to produce grounded answers with explicit citations. Across FinanceBench, Mini-Wiki, and SeaCompany, Orion-RAG delivers state-of-the-art generation quality and retrieval precision while maintaining linear scalability and strong runtime efficiency, outperforming iterative and graph-heavy baselines. The work emphasizes interpretability and HITL verifiability, offering a practical solution for enterprise data with fragmented, disconnected evidence databases.

Abstract

Retrieval-Augmented Generation (RAG) has proven effective for knowledge synthesis, yet it encounters significant challenges in practical scenarios where data is inherently discrete and fragmented. In most environments, information is distributed across isolated files like reports and logs that lack explicit links. Standard search engines process files independently, ignoring the connections between them. Furthermore, manually building Knowledge Graphs is impractical for such vast data. To bridge this gap, we present Orion-RAG. Our core insight is simple yet effective: we do not need heavy algorithms to organize this data. Instead, we use a low-complexity strategy to extract lightweight paths that naturally link related concepts. We demonstrate that this streamlined approach suffices to transform fragmented documents into semi-structured data, enabling the system to link information across different files effectively. Extensive experiments demonstrate that Orion-RAG consistently outperforms mainstream frameworks across diverse domains, supporting real-time updates and explicit Human-in-the-Loop verification with high cost-efficiency. Experiments on FinanceBench demonstrate superior precision with a 25.2% relative improvement over strong baselines.

Figures (8)

• Figure 1: System overview of Orion-RAG. (1) The Path-Annotation Data Augmentation subsystem (left) employs dual-layer labeling agents to construct hierarchical navigation paths from fragmented text, enabling real-time incremental indexing. (2) The Multi-Layer Hybrid Retrieval subsystem (right) utilizes these paths as explicit logical signposts, integrating sparse and dense search to guide the generator towards accurate and interpretable answers.
• Figure 2: Retrieval Performance Comparison. Orion-RAG demonstrates a superior balance of Hit Rate and Precision across diverse datasets.
• Figure 3: Generation Performance Comparison. Orion-RAG demonstrates superior semantic alignment and factual accuracy.
• Figure 4: HITL refinement: Injecting a specific tag reduces semantic distance, securing correct retrieval.
• Figure 5: System overview of Orion-RAG. (1) The Path-Annotation Data Augmentation subsystem (left) employs dual-layer labeling agents to construct hierarchical navigation paths from fragmented text, enabling real-time incremental indexing. (2) The Multi-Layer Hybrid Retrieval subsystem (right) utilizes these paths as explicit logical signposts, integrating sparse and dense search to guide the generator towards accurate and interpretable answers.