Neuratio: Transforming Industrial After Sales Support through Multi Agent AI and RAG - From Automation to Confidence Based Decision Making

Industrial machinery manufacturers handle thousands of after-sales support requests each day, from malfunction reports to spare-part inquiries. Responses often rely on manual searches through extensive documentation and the tacit knowledge of senior technicians. As a result, information remains fragmented, leading to slow resolution times, inconsistent service quality, and gradual loss of expertise. Existing AI solutions, such as chatbots and Retrieval-Augmented Generation (RAG) systems, have shown promise in automating customer support but remain inadequate for industrial contexts. They typically operate as generic assistants with limited domain adaptation, lack integration with company workflows, and provide no measurable confidence in their outputs. In environments where incorrect suggestions can cause costly downtime or safety issues, these limitations hinder their adoption in production settings. This thesis presents Neuratio, an AI ticketing platform developed to automate and structure industrial after-sales support. The system integrates seamlessly with existing email workflows, classifies customer requests, verifies their completeness, and generates context-aware responses using a multi-agent RAG architecture. By combining machine-specific documentation, historical tickets, and spare-part catalogs, Neuratio transforms unstructured knowledge into a reusable corporate memory. The core research contribution is a Confidence-Based Escalation Module that quantifies the reliability of generated answers by combining three complementary signals: retrieval certainty, generation consistency, and linguistic confidence. Based on this composite score, the system autonomously decides whether to reply directly, propose a draft for human validation, or escalate the case entirely.