Developing AI-Powered Chatbots as Learning Assistants to Enhance Undergraduate Computer Science and Engineering Education

Abstract

This paper presents the design, classroom deployment, and evaluation of AI-powered chatbots developed as learning assistants for undergraduate computer science and engineering education. To address growing concerns that large language model (LLM)–based tools encourage student over-reliance and hinder critical thinking, our chatbots emphasize guided reasoning over direct answer provision. Grounded in constructivist learning theory and instructional scaffolding principles, the systems foster conceptual understanding by providing structured hints, step-by-step explanations, and reflective prompts that guide student reasoning without revealing complete solutions. We developed multiple implementations using the Rasa framework and OpenAI GPT-based platforms, and conducted a mixed-methods study across three courses—an undergraduate C++ programming course, an assembly language course, and a freshman college experience course—to examine student interaction patterns, perceived usefulness, and learning outcomes. Data were collected via comparative testing, post-interaction surveys, and informal student interviews. The findings indicate that students value the guided approach, reporting higher engagement, improved confidence, and deeper understanding compared to direct-answer AI tools. However, challenges remain, particularly regarding students’ adjustment to non-solution-based assistance and natural language understanding (NLU) limitations that occasionally resulted in ambiguous guidance. This study contributes empirical evidence that guided AI dialogue serves as a pedagogically aligned alternative to direct-answer AI tutors. Furthermore, it offers practical insights into instructional integration, system design, and deployment challenges in authentic classroom settings. Future work will explore Google Vertex AI to support scalability and enhanced analytics, improve NLU accuracy, incorporate adaptive learning mechanisms, embed assessment instruments within workflows, and conduct controlled comparative studies across diverse AI platforms and course contexts.