AutoRAGTuner: Automated Optimization of RAG Pipelines

A new framework called AutoRAGTuner automates the optimization of Retrieval-Augmented Generation (RAG) pipelines, which are crucial for enhancing large language models (LLMs) but are highly sensitive to architecture and hyper-parameter configurations. The framework uses a declarative, configuration-driven approach to manage the entire RAG life cycle—construction, execution, evaluation, and optimization. It features a modular architecture that decouples pipeline stages via a component registration mechanism. To handle heterogeneous data, AutoRAGTuner introduces the Domain-Element Model (DEM), which represents objects as atomic elements with bidirectional pointers supporting nodes, edges, and hyperedges. An adaptive Bayesian optimization engine enables end-to-end hyper-parameter tuning. Experimental results show that AutoRAGTuner achieves architectural generality across diverse RAG pipelines, from vanilla to graph-based, consistently improving performance without manual tuning.