Multi-agent system architecture for secure data integration in industrial environments with multiple encryption layers

Abstract

Large-scale industrial data integration within the telecommunications production chain faces critical bottlenecks due to fragmentation across heterogeneous silos and security constraints that impose up to seven layers of encryption. This article describes the development and validation of a software architecture, called Cyber Nexus, designed to optimize Root Cause Analysis (RCA). The main objective was to reduce data access and processing latency through a decentralized Multi-Agent Systems (MAS) architecture orchestrated by Large Language Models (LLMs). The methodology employed was experimental development, using a hybrid approach that combined PMBOK® practices for governance and the Scrum framework for iterative execution. The solution was implemented with microservices in FastAPI, using in-loco Retrieval-Augmented Generation (RAG) techniques to avoid centralization in traditional Data Lakes, thus preserving industrial confidentiality. The experimental results validated the effectiveness of the solution, achieving a 90% reduction in stream-reading latency and ensuring data integrity through a Semantic Firewall with active guardrails. It is concluded that the proposed architecture establishes a new paradigm of secure interoperability for Industry 4.0.