A Retrieval-Augmented Generation (RAG) System for Supporting Architectural Design of Intelligent Transportation Systems
A Retrieval-Augmented Generation (RAG) system designed to provide automated assistance in ITS architectural design is proposed, which leverages the capabilities of Large Language Models (LLMs) while integrating knowledge from ITS reference architectures—established and validated frameworks.
Abstract
Intelligent Transportation Systems (ITS), which encompass a wide range of transportation applications leveraging communication and information technologies (e.g., sensors, identifiers, and detectors), are rapidly expanding to meet the growing demand for safer, more efficient, and sustainable mobility solutions. However, the architectural design of these systems remains challenging, particularly for designers unfamiliar with the ITS domain. To address this issue, we propose a Retrieval-Augmented Generation (RAG) system designed to provide automated assistance in ITS architectural design. Our approach leverages the capabilities of Large Language Models (LLMs) while integrating knowledge from ITS reference architectures—established and validated frameworks. As the end system's reliability depends on the relevance of the retrieved documents, this paper focuses on the retrieval stage of the system pipeline by introducing a filtering-based retrieval method. We evaluate its performance across 8 light open-source LLMs, with results indicating that Mistral Small 3 outperforms other models in retrieving relevant documents with respect to user queries. The system's provided assistance levels are exemplified.