University of North Texas
PRISM: Personalized Retrieval-Integrated System for Multimodal Adaptive Learning
Abstract
dc:descriptionTraditional educational models and their digital successors have consistently failed to resolve the tension between the diversity of learners and the uniformity of instruction, and existing AI-driven adaptive learning systems remain limited by their treatment of pedagogical decision-making as a prompting convention rather than an architectural commitment. This dissertation addresses that gap through a three-paper progressive research design grounded in Cognitive Load Theory and the Technology Acceptance Model. Paper 1 introduces NeuroQuest, a multi-agent AI framework that coordinates specialized agents to retrieve, synthesize, and deliver structured educational content in multimodal formats, with a user evaluation of seventeen participants demonstrating high satisfaction across content quality, readability, and coherence. Paper 2 introduces PRISM, the Personalized Retrieval-Integrated System for Multimodal Adaptive Learning, an agentic learning assistant that implements each pedagogical decision as a distinct auditable stage, grounds every response in retrieved evidence, and subjects every draft to internal quality evaluation before delivery. An evaluation covering 3,170 scored runs across four university courses and three learner profiles demonstrated a weighted composite score of 0.683 versus 0.406 for an LLM-only baseline (Cohen’s d = 1.16), with refusal correctness reaching 0.948 versus 0.106 (Cohen’s d = 4.58). Paper 3 validates PRISM from the student perspective through a 57-item survey administered to 85 students, in which all thirteen perceptual constructs exceeded the neutral midpoint, an OLS regression explained 80.7% of variance in Perceived Adaptive Learning Impact, and a cross-paper convergent validation confirmed that perceived cognitive effort was statistically equal across degree groups despite an objective complexity gradient spanning more than twice the range. A novel construct, Relevance and Cognitive Fit, demonstrated strong preliminary psychometric evidence (alpha = .92, r = .73). Together the three papers demonstrate that pedagogical trustworthiness can be built into an AI-driven learning system as a structural property and that students recognize and value these properties in practice.
Degree
thesis:*- Grantor dc:publisher
- University of North Texas
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Mannuru, Nishith Reddy 1998-
- Contributors dc:contributor
-
- Lund, Brady D.
- Cockerham, Deborah P.
- Nguyen, Thuan
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Public
- Mannuru, Nishith Reddy 1998-
- Copyright
- Copyright is held by the author, unless otherwise noted. All rights Reserved.
- Language dc:language
- English
Identifiers
dc:identifier.*- Identifier
-
local-cont-no: submission_4930
ark: ark:/67531/metadc2652440