A Comprehensive Educational Platform based on Generative Artificial Intelligence

Summary

Leveraging generative artificial intelligence (GAI) for teaching support, the platform delivers tailored learning experiences adaptable to users' diverse needs. It also provides abundant practical learning approaches that not only enrich the learning process but also foster lifelong learning competencies in the digital age, aligning with the goals of sustainable development in engineering education.

Abstract

Generative Artificial Intelligence (GAI) has witnessed significant progress in recent years, demonstrating transformative potential for education by enabling dynamic content creation and personalized learning pathways. However, many existing educational platforms rely on static content and fixed pathways, lacking the integrated architecture needed to fully harness GAI's capabilities to create cohesive, adaptive learning experiences. To address this gap, this study details the design, implementation, and empirical evaluation of a novel, comprehensive educational platform built upon GAI. The platform is based on a three-layer architecture transcending traditional frameworks: a Basic Technical Layer integrating modular AI models (e.g., Transformer, attention-based CNN-BiLSTM), a Processing Centre for real-time data synthesis and model optimization, and an Application and Interaction Layer housing eight core functional modules, including personalized learning, intelligent Q&A, and competency assessment. This integrated architecture facilitates a dynamically customizable learning experience that continuously adapts to individual learners' needs and progress. To evaluate the platform's efficacy, we conducted a randomized controlled trial (RCT) involving 50 undergraduate students and 20 educators over two semesters, comparing outcomes against a control group using traditional methods. Experimental results demonstrate that the proposed platform significantly improves learning outcomes, increases student engagement metrics (e.g., time-on-task and content interaction rates), and achieves high accuracy in personalized content matching. The findings suggest that this GAI-based platform constitutes an advancement in educational technology by effectively personalizing instruction and supporting adaptive learning at scale. This study contributes a detailed, replicable architectural blueprint and provides empirical evidence supporting the practical value of integrated GAI systems in enhancing educational effectiveness and fostering lifelong learning competencies.

Introduction

Amidst accelerating technological innovation, particularly in artificial intelligence (AI), tools like AI are finding increasing application in engineering education. At the same time, people have reached a consensus that the goal of promoting sustainable development is a global pursuit. Therefore, integrating sustainable development into engineering education—simultaneously improving industry technical levels and enhancing learners' sustainability awareness—has become a key focus of current education reform, driving the exploration of AI-enabled educational innovations. AI's application in education opens up possibilities for personalized learning....

Protocol

This study was conducted in accordance with the ethical guidelines of The Education University of Hong Kong. Informed consent was obtained from all participants prior to data collection.

1. Platform Specification

The evaluated system is a custom research prototype, implemented as a web-based application. The software environment is built on Python 3.9 and TensorFlow 2.12. The core AI models (CNN-BiLSTM, Transformer) are hosted on an NVIDIA A100 GPU (40 GB VRAM) within a server-based processing pipeline. The architecture consists of three layers: First, a basic technical layer integrating an attenti....

Results

The randomized controlled trial revealed significant differences between the GAI platform group (experimental) and the traditional LMS group (control). For learning gain, the experimental group scored significantly higher on the post-test (M = 85.2, SD = 6.4) than the control group (M = 76.8, SD = 7.1); mean difference = 8.4, 95% CI [4.5, 12.3], t(68) = 4.32, p < 0.001, Cohen’s d = 1.02. For engagement (time-on-task), the experimental group spent more time (M = 42.5 min, SD = 8.2) than the control group (M = 28.7 min,.......

Discussion

This study set out to test the central research hypothesis: that a GAI-enabled personalized learning platform can significantly reduce engineering learners’ cognitive load, improve learning efficiency, and enhance both ESD-related learning outcomes and overall learning experience, compared to traditional non-personalized engineering instructional models. Our randomized RCT results provide full empirical support for this hypothesis, with three key findings directly aligned to our pre-specified outcomes. First, the p.......

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
A100 GPU (40 GB VRAM)	NVIDIA Corporation	900-4H100-0000-000	Used for hosting core AI models within the server-based processing pipeline.
en_core_web_md (spaCy model)	Explosion AI	https://spacy.io/models/en#en_core_web_md	Pre-trained general academic English model used for tokenization and lemmatization. Version: v3.5.
Python	Python Software Foundation	https://www.python.org/	Programming language for the software environment. Version: 3.9.
SlimPajama dataset	Cerebras Systems Inc.	https://www.cerebras.net/blog/cerebras-releases-slimpajama-a-cleaned-version-of-redpajama/	Large, open-source pre-training corpus derived from RedPajama.
spaCy	Explosion AI	https://spacy.io/	Natural Language Processing library for text preprocessing. Version: 3.5.
TensorFlow	Google LLC	https://www.tensorflow.org/	Open-source machine learning framework. Version: 2.12.