An Adaptive AI-Driven Framework for Personalized Study Scheduling and Exam Preparation Using Locally Hosted Large Language Models

Abstract: The exponential growth of academic content and the heterogeneity of learner aptitudes have rendered fixed, one-size-fits-all study planning increasingly inadequate for contemporary students. This paper presents an adaptive, artificial-intelligence-driven framework that automatically constructs personalized study schedules and exam- preparation pathways by reasoning over individual learner profiles, topic difficulty, and proximity to assessment deadlines. The proposed system couples a Python-based scheduling engine with a Node.js presentation layer and integrates a locally hosted large language model served through Ollama, thereby preserving data privacy and eliminating recurring cloud-inference costs. A feedback-aware profiler continuously revises the learner model from quiz outcomes and study-session telemetry, enabling spaced, priority-weighted re-planning. The framework was evaluated against static, rule-based, and cloud-LLM baselines using schedule adherence, mastery progression, recommendation relevance, and inference latency as metrics. Experimental observations indicate that the proposed approach attained approximately 88% schedule adherence and a 26-percentage-point gain in average mastery over an eight-week horizon relative to a static baseline, while sustaining acceptable local-inference latency. The principal contributions are a privacy-preserving on-device intelligence layer, an adaptive re-scheduling algorithm that fuses forgetting-curve and difficulty signals, and an integrated assessment loop that closes the gap between planning and measured learning outcomes.

Keywords: Personalized learning; adaptive scheduling; large language models; Ollama; educational technology; spaced repetition; on-device inference; intelligent tutoring systems.