ToriLearn: An AI Enabled Adaptive Learning System for GCE Chemistry Preparation

International audience Secondary school students preparing for the General Certificate of Education examinations in Cameroon face a documented 62 percent failure rate in core science subjects, alongside heavy reliance on static revision material that does not adjust to what a learner already knows. This paper presents the design and implementation of ToriLearn, a prototype web based adaptive learning platform built to address this gap for GCE Ordinary Level Chemistry. The system combines a React and TypeScript frontend with a Django REST Framework backend organised into four layers, namely presentation, API, application logic, and data persistence. Within the application layer, a Bayesian Knowledge Tracing engine estimates mastery of eleven Chemistry knowledge components in real time, a Gradient Boosting classifier recommends chapter level progression, a K-Means clustering module groups learners by behaviour, and a conversa-tional tutor built on Groq hosted large language models supplies explanations with an offline fallback chain. The system was built across five Agile Scrum sprints and evaluated through functional testing, performance benchmarking on a local development server, and mathematical simulation of three learner trajectories. Results show that all core modules function correctly, that the adaptive quiz endpoint responds within 45 to 80 milliseconds under development conditions, and that the simulated learner scenarios behave in a manner consistent with the underlying Bayesian model. No live classroom pilot was conducted, so claims in this paper are limited to architectural soundness and simulated behaviour rather than measured learning gains. We discuss how this layered, fallback aware design compares with existing intelligent tutoring architectures and outline what a subsequent classroom deployment would need to confirm.