This project explored the integration of multiple AI platforms into the university’s virtual learning environment (Learning Mall) to deliver asynchronous language and academic skills support for second-year EAP students studying with the School of Advanced Technology. The design combined generative AI tools, gamified H5P activities, and two customized chatbots: one academic-focused (“Cork the Wisdom Bot”) and one conversational (“Sara the Classmate Bot”). These components were embedded into a single page to create an interactive, multimodal learning experience. XIPUAI and other platforms were employed to extract course content, generate quiz items, design themed narratives, and support aesthetic choices. The project demonstrated how AI can streamline materials creation, enhance learner autonomy, and promote regular engagement beyond class time. Limitations, such as platform restrictions and the need for human oversight, were also identified. This initiative provides a replicable model for embedding AI-driven language support into digital learning ecosystems, and future studies may further examine learning outcomes, student engagement, and AI literacy development.

AI in education; asynchronous learning; H5P; chatbot; gamification; virtual learning environment (VLE); English for Academic Purposes (EAP); autonomous learning; generative AI; student engagement

 

The increased accessibility and diversity of AI tools provide unique opportunities for creating interactive asynchronous learning options that can encourage students to engage with course content outside of the classroom, practice their language skills and become more adept at interacting with artificial intelligence.

Asynchronous learning now frequently involves the use of digital materials to extend learning outside of the classroom and class time and has been shown to greatly benefit students, notably in the field of language learning (Sulha, Famela and Harahap, 2021; Dayag & Faramarzi, 2024). Virtual Learning Environments (VLEs) allow digital materials to be presented to students in increasingly diverse ways and have been conclusively proven to enhance both cognitive and non-cognitive learning across a range of student levels (Jing et al., 2024). Amongst the many possibilities that these platforms offer, H5P, short for HTML-5-Packag,  is a particularly adaptable and convenient way to create and deliver interactive materials including games, quizzes and audio, with Doran and Briggs (2024) recording high levels of effectiveness and student engagement using these materials with Chinese university students, integrating them directly into the Learning Mall VLE. Also increasing in popularity is the integration of Artificial Intelligence into pedagogy, in the form of Chatbots customized to fit specific educational purposes and assist in materials design.

AI chatbots can now be designed as incredibly nuanced and effective learning and teaching partners, both for conversation and the exploration of learning materials (Wu & Yu, 2024). Gökçearslan, Tosun & Erdemir (2024) reviewed over 30 recent articles to conclude that while much of the research and guidelines published have focused on ChatGPT, it is now apparent that the age of AI in education has dawned, and a plethora of new tools are emerging almost daily. The same review noted increases in learning motivation and language skills as the main advantages for this type of intervention. While some may suspect the novelty aspect of such a new technology might be influencing these findings, longer term studies are already producing robust data on improved learning outcomes and reduced learning anxiety across a variety of educational contexts (Wu & Yu, 2024). Educators also noted the time saving aspect of AI which allows for rapid extraction of key learning points from a syllabus and can assist in a number of materials generation tasks (Gökçearslan, Tosun & Erdemir, 2024).

The objective of the current project was to explore the aforementioned developments in digital materials, VLEs, and AI agents can be combined to supercharge asynchronous language and content support. The project was developed using the Learning Mall VLE platform which provides versatile page setup options using HTML to integrate several interactive sections delivering asynchronous language and course content support for year two English for Academic Purposes (EAP) students from the school of advanced technology. This revolves around four key elements. The first is using AI to extract language, conceptual content and important course-related information from a number of course documents such as workbooks, the syllabus and assessment task sheets. This information was then used to create several learning sections. One was a custom AI agent using XJTLU’s own XIPU AI platform, with a specialised knowledge base made up of the content of the Semester 2 School of Languages, EAP111 course, English Language & Study Skills for Advanced Technology. The principal function is to answer any queries students may have about the assessments, deadlines, key concepts and vocabulary from the course. The same information was also used to create an interactive gamified learning area where students can test and improve their knowledge via s series of fun activities organised into a themed Game Map. Finally, a second custom AI agent (Doubao platform) was created to serve as an English language partner to practice and improve less formal conversational English skills. These different sections were integrated into a single functional and visually appealing page on Learning Mall.

This was developed to provide students with interactive, weekly opportunities to reinforce language and literacy skills from the EAP111 course. Combining generative AI with H5P tools enabled the development of asynchronous learning materials that encouraged self-assessment, gamified revision, and cognitive engagement (Doran & Briggs, 2024; Homanová & Havlásková, 2019; Reyna, Hanham & Todd, 2020).

Firstly, EAP111’s Module Handbook, Syllabus Outline, and weekly workbooks were analysed to extract relevant linguistic input and literacy tasks aligned with course outcomes. Drawing on recent findings which highlight the effectiveness of H5P for flipped and interactive learning environments (Doran & Briggs, 2024), the Game Map tool was selected as the primary interface. The H5P Game Map enables integration of multiple activity types, including multiple choice quizzes, drag-and-drop tasks, matching games, and audio-visual materials, into themed hotspots, facilitating an immersive and learner-driven experience (Homanová & Havlásková, 2019). 

To contextualize the activities within the weekly course themes, generative AI (ChatGPT and XIPUAI) was employed to assist in creating narratives and interactive structures. For example, Week 2, focused on autonomous public transportation, was transformed into a smart city scenario, where students, acting as junior engineers, explore transport hubs and complete tasks to optimise the system. This narrative approach was chosen to increase learner motivation through role-based participation and task progression (Reyna, Hanham & Todd, 2020; Homanová & Havlásková, 2019; Doran & Briggs, 2024).

Quiz questions and task instructions were generated using AI to draft content based on the course materials. All content was then manually edited to align with course objectives, ensure CEFR-appropriate difficulty, and maintain instructional clarity. This process highlights the utility of AI in accelerating initial content creation, while reaffirming the necessity of human oversight to ensure clarity, contextual relevance, and instructional effectiveness (Zirar, 2023; Wrigley, 2018).

Visual design was supported by royalty-free AI-generated imagery sourced from Freepik to create the game map background (Figure 1). This contributed to the immersive feel of the interface, while ensuring accessibility and ease of distribution (Aktay, 2022). Attention was also given to navigation, visual clarity, and device compatibility to minimise cognitive load and support autonomous learning (Wilkie et al., 2018; Reyna, Hanham & Todd, 2020).

 

Fig. 1 Gamified self-testing area

 

Wisdom Bot is a specialized AI agent with a knowledge base made up of the content of semester 2 of the EAP111 course. The principal function is to answer any questions students have about assessments, deadlines and weekly course content, such as key concepts and vocabulary. The bot was designed to respond in a way that encourages students to read the materials, such as by referring to specific workbooks or documents, and complete the activities in the Gamified Self-Testing Area. 

The first step was to create the knowledge base on XiPU AI. A knowledge base, as defined by Holmes and Stocking (2009, p.6) is “the persistent collection of knowledge that supports the operation of an intelligent system.” This comprehensive collection of statements is used to define and model the environment in which the agent operates (Holmes and Stocking, 2009). Therefore, to create the knowledge base, essential documents, such as task sheets, syllabus and teaching plan, and weekly workbooks were uploaded, totaling 12 documents in all. Once compiled and published, a link and knowledge base ID number was created in order to incorporate it into the Chatbot. For the chatbot, there were two available options: an AI block available through LMO or creating a custom Agent on XIPU AI. Using the integrated AI block on LMO allows a ChatGPT knowledge base with the added feature of precise course specifications added through the “Source of Truth” function. However, a XIPU Agent allows you to only use a knowledge base of your own creation.  Ultimately, the AI block on LMO seemed to be the better option because it could be more seamlessly integrated onto the course page, albeit in the sidebar. For the XIPU agent, the only option was to share a link or QR code, so students were limited to accessing the chatbot externally.

 

 

Figure 2: Configuring Wisdom Bot -  Image from Learning Mall webpage. 

 

To integrate the Chatbot, a XIPU AI chat block was added on LMO. As can be seen in Fig. 2, the block was given a working title of “Wisdom Bot” and set on Knowledge Base mode. The next entry required is the Source of Truth which guarantees that the chatbot provides the same answer to similar queries, and it requires that information be in a question-and-answer format. For example, “Q: When is section 3 due? A: Thursday, March 16.” This section is particularly useful if you are using the block in ChatGPT mode in order to ensure the information provided is accurate and consistent. To streamline the question creation process, the essential course documents, such as task sheets, syllabus and teaching plan, handbook and weekly workbooks were uploaded into DeepSeek, which is an AI language model that uses reinforcement learning and model distillation (Okaiyeto et al, 2025). It was then asked to “Write 50 questions that students may ask based on the information provided in a Q and A format.” It quickly created questions under 8 subheadings: General, Portfolio Tasks, Report Structure, Technology Topics, Grading and Penalties, Speaking Coursework, Final Exam and Miscellaneous. After double checking that all Q and As were appropriate and correct, they were copied and pasted into the Source of Truth box on LMO. 

The final setting is the completion prompt which is the prompt the AI will be given before the conversation transcript. A user can influence the AI’s “personality” based on the description. Our completion prompt stated: “Generally, you are a EAP111 teacher. You are helpful, easy-going, and encouraging, and you should direct students to specific documents to read for themselves. You are a teacher who sees your students twice a week and are here to answer additional questions. You may start by asking the student what they need help with today. The purpose is to help students find information and tell them which specific documents to read for more information.” After completing all the settings and information, the Knowledge Base ID that was generated at the start was entered under Knowledge Base mode, as seen in Fig. 2.

The Doubao AI platform, developed by Byte Dance, offers several advantages in the field of artificial intelligence (Zhang, 2024). Doubao’s LLM model has been trained using Byte Dance’s self-developed data tagging system, absorbing extensive knowledge from various fields, including history, science, technology, culture, and more (Ai Faner, 2025). This capacity distinguishes a customized agent in Doubao from other agents derived from models like GPT, Claude or other mainstream LLMs. 
The web version of Doubao’s toolbar does not include AI Agent creation function, so users must download and install the software version to custom an AI Agent. Within this software version’s toolbar, in the right column, there are two tools for creating an agent: the AI Agent Generator and the AI Agent. Users who are uncertain about prompting or lack a clear customization plan for the agent can opt for the AI Agent Generator.

In the AI Agent Generator tool, step-by-step instructions are provided to gather details about the agent’s role, personality, educational background, hobbies and avatar, culminating in the creation of a classmate bot. The Classmate Bot (Sara) developed for this project is designed as a fun and informal foreign classmate studying at XJTLU in the School of Advanced Technology. Sara is a computer science major who prefers to speak only in English, has good general and topical knowledge about the EAP course, and excels at assisting students with language challenges while engaging in conversations about class-related topics and her hobbies, including video games, classic movies, and badminton. Such productive speaking skills are prioritized as there is a strong emphasis on applying English language skills for authentic discussions in EAP learning and teaching (XJTLU, 2025).

Once the avatar and language were chosen for the Classmate Bot, the Bot was finalized with an avatar, which contained a range of voices with different personalities, pitch and speaking speed. In addition, users can click "Clone My Voice" to create personalized voice packs by recording their own voices. When Classmate Bot (Sara) was customized, it underwent testing through several conversations ranging from basic greetings to general topics on movies and books (Figure 3). It responds in both text and audio formats, adjusting its replies based on user feedback in a conversational manner. As illustrated in Figure 3, Classmate Bot (Sara) responds to questions on a variety of topics, from discussions about badminton rackets to autonomous transportation, in various lengths while asking human-like follow-up questions to maintain the conversation flow.

 

 

Figure 3. Classmate Bot in Doubao AI Platform

 

Although the current Classmate Bot appears to be functioning effectively, one limitation of the Doubao platform is that users cannot build a knowledge base or upload materials to tailor the agent further. Ultimately, this agent is expected to serve as a language partner for second-year computer science students, enabling them to practice their conversational English using text or audio while providing feedback on their language use. 

Learning Mall VLE integration

There were now three sections: Cork (Wisdom Bot), Sara (Classmate bot) and the Gamified Self-Testing Area. The next step was to integrate them into a single Learning Mall page which would be functionally easy to use and visually appealing to students. 

Kronemann, et. Al (2023) reviewed marketing literature to conclude that, although contested by some, there is evidence that AI provokes much greater interaction when it presents humanlike features and behaviors, particularly asking questions and sharing opinions or ideas. It was therefore decided that both the AI bots would have an animated human “face” to match their pre-programmed identities, for example Sara looks the part of a friendly student of about 20 years old while Cork has the appearance of an older teacher wearing glasses. XIPU and Sora were utilized to generate these images as animated GIFs and embedded them in the page, to give each area an identity and bring the AI to life.

For the functionality part, the page was kept very simple and clear: students can interact directly via a text box on the LMO page and receive responses there. For Doubao, students can go a step further and download the app, allowing textual and audio communication on their phone.

To summarize, the project provided essential experimentation for educators to experience integrating these increasingly common AI tools and delivering asynchronous language and content support. At every stage AI was used to streamline and simplify the process of extracting content and language from the course materials and transforming it into an engaging digital format. AI also featured heavily in the aesthetic side of the materials design, providing design ideas, images, videos and more. Once created, these materials can be accessed multiple times without the need for additional intervention from the educator.

The final result was to bring the course to life and enable students to interact with the learning process in a variety of ways depending on their needs and mood. The Wisdom Bot provides reliable support for serious academic content that could affect the students’ assessment performance. The Gamified Self-Testing materials provide a fun and engaging way to check and reinforce understanding. And the Classmate Bot has the potential to promote natural language practice in a less formal way, increasing student language ability without the pressure or constraints of formal materials. The learning process becomes more autonomous in every way. Future research would of course be required to quantify the effectiveness of these types of AI-driven materials, looking for example at language ability, content knowledge and the amount of time spent engaging with the materials and agents.

Another important goal of the project was to design a system that would increase student familiarity in interacting with AI systems, something which is undoubtably becoming essential in academic and professional life. Further study could also test this aspect of the project.

 

 

 

Ai Faner (2025 April 10). Doubao releases the large model 1.5 Pro, and after testing, I found that this domestic AI defeated GPT-4 using the most 'foolish' method. Tencent Webpage. https://news.qq.com/rain/a/20250320A04KCH00
Aktay, S. (2022). The usability of images generated by artificial intelligence (AI) in education. International Technology and Education Journal, 6(2), 51–62. https://dergipark.org.tr/en/download/article-file/2888675
Dayag, Joseph & Faramarzi, Sajad. (2024). Virtual Learning Environment: Enriching 21 st Century Academic Landscape. European Journal of Theoretical and Applied Sciences. 2. 248-253. 10.59324/ejtas.2024.2(1).19.
Doran, S., & Briggs, C. (2024). H5P Interactive Materials for Flipped Learning: Engagement, Effectiveness and Challenges. In ThaiTESOL International Conference Proceedings, October 2024. p.p. 32 - 47 
Doran, S. (2020). Using asynchronous online discussions to promote out of class learning in Vietnamese University students. Asian EFL Journal, 27(2.3) p188-211 https://www.asian-efl-journal.com/monthly-editions-new/page/3/index.htm 
http://eprints.unm.ac.id/21569/1/Jurnal%20International%20bereputasi%202020-Hikmah-Basri-Amirullah-Asian%20EFL.pdf
Gökçearslan, S., Tosun, C., & Erdemir, Z. G. (2024). Benefits, challenges, and methods of artificial intelligence (AI) chatbots in education: A systematic literature review. International Journal of Technology in Education, 7(1), 19-39.
Jing, Y., Dai, J., Wang, C., Shen, S., & Shadiev, R. (2024). Unleashing the power of virtual learning environment: exploring the impact on learning outcomes through a meta-analysis. Interactive Learning Environments, 33(1), 52–69. https://doi.org/10.1080/10494820.2024.2335481
Homanová, Z., & Havlásková, T. (2019). H5P interactive didactic tools in education. In EDULEARN19 Proceedings (pp. 9266–9275). IATED.
Kronemann, B., Kizgin, H., Rana, N., & K. Dwivedi, Y. (2023). How AI encourages consumers to share their secrets? The role of anthropomorphism, personalisation, and privacy concerns and avenues for future research. Spanish Journal of Marketing-ESIC, 27(1), 3-19.
Okaiyeto, S. A., Junwen Bai, Jun Wang, Mujumdar, A. S., & Hongwei Xiao. (2025). Success of DeepSeek and potential benefits of free access to AI for global-scale use. International Journal of Agricultural & Biological Engineering, 18(1), 304–306. https://doi.org/10.25165/j.ijabe.20251801.9733
Reyna, J., Hanham, J., & Todd, B. (2020). Flipping the classroom in first-year science students using H5P modules. In EdMedia+ Innovate Learning (pp. 1077–1083). Association for the Advancement of Computing in Education (AACE).
Sulha, A. H., Famela, A. K., & Harahap, A. T. A. (2021). The implementation of synchronous and asynchronous learning in English as foreign language setting. Etdc: Indonesian Journal of Research and Educational Review, 1(1), 17-27.
Wilkie, S., Manwaring, P., Zakaria, G., & Chen, L. (2018). Implementing multimedia learning principles and visual design to improve learning outcomes. ASCILITE Proceedings, 344–355.
Wrigley, T. (2018). The power of ‘evidence’: Reliable science or a set of blunt tools? British Educational Research Journal, 44(3), 359–376. https://doi.org/10.1002/berj.3338
Wu, R., & Yu, Z. (2024). Do AI chatbots improve students learning outcomes? Evidence from a meta‐analysis. British Journal of Educational Technology, 55(1), 10-33.
XJTLU (2025, May 16). Language Teaching. https://www.xjtlu.edu.cn/en/study/departments/school-of-languages/english-language-centre/language-teaching/eap-year-one
Zhang, J. (2024). TikTok Parent ByteDance Now Has China’s Most Popular AI Chatbot. Bloomberg.Com. 
https://eds.p.ebscohost.com/eds/detail/detail?vid=0&sid=86d621fe-c68c-4a46-844b-56dc7d0dd58b%40redis&bdata=JnNpdGU9ZWRzLWxpdmUmc2NvcGU9c2l0ZQ%3d%3d#AN=177356009&db=bsu
Zirar, A. (2023). Exploring the impact of language models, such as ChatGPT, on student learning and assessment. Review of Education, 11, e3433. https://doi.org/10.1002/rev3.3433