Abstract:

Standardized comprehensive examinations remain a critical benchmark of academic quality and professional readiness in engineering education. However, students enrolled in challenge-based and competency-driven educational models often demonstrate lower-than-expected performance on high-stakes, multiple-choice standardized tests, not due to a lack of knowledge but rather to limited structured exam preparation. This study presents the design, implementation, and evaluation of an AI-powered adaptive learning agent developed to enhance standardized exam performance among senior Industrial Engineering students.

The proposed system integrates a large language model–based conversational agent that communicates with students via email and Telegram, enabling them to generate unlimited practice questions aligned with the national Mexican exit examination framework (CENEVAL). The agent dynamically adapts question difficulty and topic distribution based on individual performance, provides immediate formative feedback with explanatory reasoning, and records longitudinal performance data in a structured learning analytics dashboard. The instructional strategy is grounded in adaptive learning principles, formative assessment theory, and data-driven personalization.

A quasi-experimental design was implemented across two academic cohorts. The 2023, 2024, and 2025 graduating classes served as the control group, whereas the 2026 cohort used the AI agent as an intervention. Key variables measured include percentage of correct responses over time, total interaction time with the agent, and official standardized exam scores. Statistical analyses (t-tests and Pearson/Spearman correlations) are used to determine the relationship between agent usage and performance improvement.

Preliminary expectations project an increase in the proportion of students achieving “Outstanding” performance and a reduction in unsatisfactory results. The innovation aims to shift student behavior toward self-regulated, data-informed exam preparation. By providing real-time feedback and progress tracking, the agent promotes metacognitive awareness and ownership of learning.

From an institutional perspective, the system reduces faculty workload in item generation and monitoring while offering a scalable, transferable architecture adaptable to other disciplines and standardized assessments. The project contributes to emerging research on AI-based adaptive learning agents in higher education, particularly in high-stakes assessment contexts. This research supports the integration of AI-powered adaptive systems as a viable, scalable strategy to enhance exam readiness, equity, and performance in engineering education and beyond.

Keywords:

Artificial Intelligence in Higher Education, Adaptive Learning Systems, Standarized Exam Preparation, Learning Analytics, Conversational AI Agents.