An AI-Powered Teaching Performance Evaluation System: Technical Implementation and Pilot Testing at Universidad Técnica de Ambato

Abstract

The article detailed the design, implementation and pilot test of an artificial intelligence-based system for evaluating teaching performance in higher education. The researcher aimed to create an automatic evaluation process by using speech recognition, semantic analysis and emotional detection, within an artificial intelligence architecture developed at Universidad Técnica de Ambato. The process was executed using open-source tools such as AssemblyAI, GPT-4o-mini, DeepFace and the n8n workflow platform which allowed autonomous analysis of recorded classroom sessions. A quasi experimental validation was actioned using 36 class recordings from 18 teachers from three disciplines. Overall, the findings indicated transcription accuracy of 96.4 %, inter-rater reliability above 90 % rubric agreement and substantial agreement with human raters (Cohen’s κ > 0.65; ICC > 0.80). Time for evaluation was reduced by greater 95 % and cost by 97 % compared with other peer review methods. These results confirmed the feasibility and reliability of the system for institutional quality assurance in teaching evaluation. The study concluded that artificial intelligence-based approaches could provide institutions with an open, efficient and scalable mechanism to assess the pedagogical performance that enhances innovation in higher education.