<![CDATA[Operational issues in digital attendance systems such as recognition errors, network disruptions, or forgotten credentials are still frequently reported through informal channels, resulting in unstructured records that are hard to trace and prone to duplication or fraud. This study proposes a conversational, report-driven attendance incident system that combines a multi-agent approach with a finite state machine (FSM) to keep dialogues structured, preserve conversational context, and ensure data completeness. The system performs automatic information extraction from user inputs using a large language model, and conducts evidence pre-validation via optical character recognition and multimodal classification prior to human verification. The research methods include organizational needs analysis, design of the data model and interaction flow, formulation of FSM rules to govern dialogue stages, and integration of conversational intelligence, information extraction, and evidence validation components. Evaluation follows User Acceptance Testing (UAT) grounded in realistic business workflows at PT. XYZ, focusing on three aspects: (1) time efficiency compared to the manual reporting process, (2) data consistency and completeness by comparing chatbot variants with and without FSM control, and (3) reduction of initial manual workload for administrators during verification. Experimental results show the system accelerates reporting by ≥90% relative to the manual method, yields higher completeness with the FSM-controlled chatbot (approximately 96% complete versus ~70% without FSM), and achieves an ≈70% reduction in early administrative intervention. Evidence validation using text recognition and multimodal classification attains ~90–95% accuracy for common cases, effectively serving as a pre-screen before human review. These findings indicate that a multi-agent approach with FSM control can improve the speed, traceability, and reliability of attendance incident reporting. ]]>
Copyrights © 2026
