BellTeus: Call Center Performance Evaluation System Using the Large Language Model

Call centers face increasing challenges in evaluating agent performance, detecting operational risks and monitoring indicators such as customer satisfaction. Traditional methods, based on human supervision, are costly, not very scalable and subjective, making traceability and continuous improvement difficult. In order to meet these needs, BellTeus was developed, an automated system that applies LLM techniques to analyze weekly historical call records, through automatic transcription, semantic analysis and normative evaluation. BellTeus converts stored audios into structured reports with key performance metrics. Its modular design allows the system to be easily scaled and integrated into different organizational environments. Experiments showed that GPT-4 is the model that obtained 88.5% accuracy in interpreting call content, identifying emotional and compliance issues from audio and metadata analysis. These findings reinforce the potential of LLMs in retrospective conversational processing within critical call center operations.