Adaptive Artificial Intelligence Learning Platform With Dynamic Roadmaps: An Architectural Proof-of-Concept

Abstract

Conventional Learning Management Systems (LMS) force all students through a single, static instructional sequence regardless of prior knowledge or learning pace, causing student disengagement and elevated attrition rates. While predictive AI approaches dynamically re-route static materials, they are constrained by fixed content repositories. This paper presents an architectural proof-of-concept for an adaptive web-based platform that utilizes Generative Artificial Intelligence (GenAI) to construct, evaluate, and re-author personalized educational roadmaps in real time. The platform integrates a diagnostic pre-test with response integrity filtering, an 80% mastery cri- terion gate, and a two-stage remediation protocol (Tier-1 micro simplification and Tier-2 macro-syllabus reconstruction). Built on Next.js 14, PostgreSQL, Redis, and BullMQ, the system implements a dual-model LLM failover architecture (Gemini 3.1 Flash primary with Claude 4.5 Haiku fallback) to ensure reliability. In an empirical evaluation of 13 filtered learner sessions, 11 sessions achieved positive cognitive growth measured by Hake’s Normalized Gain (μ = 0.674, SD = 0.321), yielding an 84.6% progression rate. While preliminary findings confirm functional viability, we acknowledge limitations including sample size constraints, absence of a control group, and potential Hawthorne effects. We outline enterprise integration pathways via IMS LTI v1.3, cognitive load fatigue mitigation, and a framework for future randomized controlled trials. Index Terms—Adaptive learning, artificial intelligence, dynamic roadmaps, learning management systems, mastery learning, remediation protocols, zone of proximal development, educational technology.