LitterBox: A Teaching Assistant ChatBot for Computer Organization and Architecture

This paper presents LitterBox, a conversational AI assistant designed to enhance student learning in Computer Systems Architecture (CSA) education through guided exploration rather than direct answers. Implemented using a Retrieval Augmented Generation (RAG) architecture, LitterBox leverages professor-provided materials, maintains conversation memory, and employs dynamic educational scaffolding that adapts to student queries. We conducted a comparative study between controlled (n=36) and uncontrolled (n=43) student groups to evaluate its effectiveness. Results demonstrate that while both groups showed similar proficiency in factual knowledge (89.8% vs. 87.2% on multiple-choice assessments), LitterBox users exhibited significantly better conceptual understanding in essay responses (3.25/5 vs. 2.88/5 for Essay 1; 3.60/5 vs. 3.31/5 for Essay 2). Notably, the controlled group showed more consistent quality with fewer below-average responses and better technical precision. LitterBox users engaged more deeply with the material, spending approximately twice as long (18:17 vs. 8:56 minutes) and averaging 7.3 follow-up questions per session. The system achieved good usability ratings (SUS score: 76.5), with students particularly valuing its conversational nature and contextual awareness. This research demonstrates how AI can ethically enhance technical education by promoting deeper conceptual understanding while maintaining academic integrity.