The rapid development of generative artificial intelligence is reshaping educational practices by enabling new forms of adaptive, scalable, and personalized learning. However, many existing AI-based learning systems remain fragmented, insuficiently integrated, and often detached from pedagogical frameworks. This paper presents the mISLec4EDU, an AI-enabled platform designed to support asynchronous and personalized learning for students while assisting educators in the design and delivering instructional content. The platform integrates a conversational tutor, adaptive formative assessments, and human-in-the-loop support for quiz creation. It also supports deployment flexibility through both cloud-based and locally hosted large language models, addressing institutional needs for privacy, cost control, and flexibility. We describe the system architecture and key functionalities, and illustrate their use through a real-world case study. The unified architecture enables context-aware prompting, learner-specific feedback, and lecturer-controlled content creation. The platform's design emphasizes pedagogical alignment, educator control, and responsible AI integration. This contribution aims to initiate discussion on practical, flexible, and ethically grounded uses of generative AI in education.
CEUR
ceur-ws.org are mathematical reconstructions of learned patterns. For educators, GenAI presents opportunities to enhance learning experiences through creative tools. Yet it also demands careful consideration of its limitations, including potential inaccuracies, biases, and ethical concerns regarding originality and ownership of generated content.
The educational use of GenAI is rapidly expanding. Several national education ministries, such as Spain and Australia, have issued guidelines promoting the pedagogically responsible and inclusive use of GenAI [4, 5]. At the same time, systematic reviews have noted that many existing applications remain fragmented—focusing either on chatbots or assessments without cohesive integration or educator oversight [6, 7].Research and development in GenAI focus on improving these models’ speed, capabilities, and eficiency. However, many fundamental questions about its principles, applications, and socioeconomic impact remain unexplored. This poses a challenge, as a clear understanding and framework for GenAI in education is still evolving. While GenAI ofers numerous opportunities for innovation and improvement across various industries, it is equally important to acknowledge its challenges and strive for its responsible and ethical use.
The need for AI regulation is particularly pressing in education due to the field’s inherent vulnerabilities. These include handling sensitive student data, the risk of algorithmic bias afecting educational outcomes, and concerns about academic integrity arising from the misuse of GenAI tools, such as essay generators. Unlike other advanced AI systems, GenAI strives to replicate a broad range and complexity of human cognitive abilities, enabling machines to perform various intellectual tasks [8]. In this way, GenAI holds exceptional potential in education, particularly in building adaptive learning environments tailored to the diverse needs of students. It facilitates dynamic adjustments to students’ abilities and learning needs comparable to human interaction.
However, GenAI solutions’ high capabilities, combined with their lack of awareness and understanding of broader contexts, as well as their uncritical use, pose several potential challenges. These include academic dishonesty, over-reliance on automated tools, the spread of biased or inaccurate information, and a decline in critical thinking and creativity among students and educators. Moreover, many platforms lack transparency, do not involve educators in the content-generation loop, and may reinforce existing inequalities [9]. By establishing clear frameworks and guidelines, appropriate regulations can mitigate these risks while promoting the ethical and efective use of AI technologies throughout the educational system.
GenAI holds significant potential to personalize education and scale the learning by dynamically adjusting to to individual learners’ needs, pace, and preferences [10]. It extends the boundaries of creativity and innovation through interactive, innovative, and often highly authentic learning experiences. These technologies can close achievement gaps, open access to high-quality education to everyone, and provide new tools to those who seek to transform pedagogic methods.
Recently, there has been a trend of emerging studies and publications covering GenAI in diferent educational contexts. North Carolina Department of Public Instruction report [11] points out GenAI’s vast aid to educators in carrying out teaching-related tasks, particularly those targeting improved teaching eficiency and efectiveness. Similarly, the Spanish Ministry of Education [ 4] pointed out that GenAI facilitates adaptive, personalized learning by tailoring lessons, exercises, and the pace of content progression to the individual learner.
Even though GenAI can be beneficial to learners and educators, let’s focus on the latter first. GenAI can be a helpful tool for them when creating content, such as developing lesson plans, quizzes, or multimedia materials, and constructing interactive learning materials. The other important use could be task automation [4]. GenAI can automate routine or repetitive tasks, such as rapid large-scale feedback generation, report writing, and communication with parents [12]. This support reduces the administrative burden on educators, allowing them to devote more time and energy to their core pedagogical responsibilities.
On the other hand, learners may also benefit from GenAI. It has the potential to make learning more
scalable by tailoring various tools and features to meet diverse students’ requirements and make the
learning process easier. It functions as a learning assistant, providing real-time support to learners across
multiple tasks, from answering questions to explaining complex or dificult-to-understand concepts [ 11].
Moreover, GenAI can increase student engagement through game-based learning, interactive tools, and
real-time feedback, making the learning experience more motivating and dynamic [
When used appropriately, GenAI improves accessibility by helping to overcome linguistic barriers and other limitations (e.g., those faced by students with special needs) [13]. This includes features such as translation, adaptive interfaces, and assistive technologies like text-to-speech and speech-to-text [5]. It can thus provide stronger support for diversity and equity, fostering the inclusion of culturally and linguistically diverse learners.
Despite its promise, GenAI also raises significant concerns. Critical literature points to the risks of generation of inaccurate or fabricated content, embedded algorithmic biases, limited transparency in how outputs are produced, and the risk of diminishing students’ autonomy in the learning process. [14, 6, 15]. For instance, AI-generated explanations can sometimes include inaccurate or misleading statements, which may go unnoticed in low-stakes contexts [16]. Over-reliance on GenAI may reduce learners’ critical thinking and problem-solving development, especially if AI is treated as a source of unquestioned authority. Furthermore, ethical concerns about intellectual property, academic honesty, and equity must be considered when deploying such technologies in diverse educational settings.
Current GenAI tools are often fragmented, requiring educators and students to use multiple applications to accomplish tasks such as content generation, tutoring, and assessment. This fragmentation increases complexity and can reduce overall efectiveness. While several tools support personalization in isolated areas, few integrate this capability into a unified, pedagogically coherent system. In response to this need, we developed mISLec4EDU, a platform designed to support both learners and educators by consolidating core GenAI functionalities into a single environment, guided by principles of responsible use and instructional alignment. 3. Case study: mISLec4EDU mISLec4EDU is an AI-enabled platform for the deployment of asynchronous and personalized learning in real-world educational settings ranging from higher education to K-12, wherever asynchronous and personalized learning is needed. Although traditional e-learning tools often provide content delivery or limited adaptivity, they do not ofer the full spectrum of personalization. Many current implementations are fragmented – for example, one tool might ofer an AI tutor chatbot, while another separately provides adaptive quizzes, leaving it to educators or learners to stitch together multiple tools. The contribution of mISLec4EDU is to unify these capabilities in one platform with a coherent architecture. The goal of such a unified approach is to enhance user experience (reducing the cognitive load from switching tools) and improve learning outcomes by ensuring that every aspect of the learning process, from content to support to assessment, is personalized and data-informed. mISLec4EDU not only focuses on the student but also acts as a supporting system for the lecturer, helping the lecturer to prepare augmented learning materials and diferent types of student knowledge evaluations, such as practices and exams.
Figure 1 provides a comprehensive overview of the primary mISLec4EDU functionalities in the form of a use case diagram. The use case diagram illustrates how students can apply for and view courses, engage with course content, practice and take exams for courses, and obtain AI-generated explanations and augmented materials. They also interact with an AI tutor chatbot capable of conversational support and tailored responses. For lecturers, the platform provides tools to manage course content, including enabling the lecturer to provide additional learning resources in the form of PDFs, create and validate assessments such as practices and exams using AI assistance, and track student progress. Both student and lecturer interactions leverage AI features that are backed by a modular integration with large language models (LLMs).
A distinguishing design principle of mISLec4EDU is its capability to utilize both third-party LLM APIs and local LLM deployments. API-based integration allows the platform to take advantage of state-of-the-art models (e.g., OpenAI’s GPT-4, Claude, or Gemini). However, dependence on external services can introduce concerns regarding data privacy, network reliability, and cost, particularly in educational institutions with strict data governance policies.
To mitigate these challenges, mISLec4EDU supports local deployment of LLMs. Running LLMs on-premise ensures that sensitive student data remains within the institutional infrastructure, avoiding transmission to external servers and reducing exposure to third-party data processing. This configuration allows educational institutions to retain full control over the data flow—from input prompts and learning context to generated outputs—ensuring compliance with data protection regulations and internal governance policies. Local deployment also enables system functionality in ofline or bandwidth-constrained environments, ensuring uninterrupted access in regions with limited internet connectivity.
A key advantage of local LLMs lies in their adaptability: models can be fine-tuned or contextualized using institution-specific educational content, which improves relevance, interpretability, and alignment with curricular goals. However, deploying and maintaining LLMs locally presents significant technical and operational demands. It requires access to specialized hardware (e.g., high-memory GPUs), robust infrastructure for inference serving, and skilled personnel to manage updates, scaling, and fault tolerance. Additionally, institutions must ensure that user access and usage data are securely handled and logged in accordance with local privacy policies.
By supporting both cloud-based and local LLM integration, mISLec4EDU provides a flexible architecture that allows institutions to choose the deployment model best suited to their privacy requirements, technical capacity, and pedagogical goals—without sacrificing the capabilities of generative AI.
The platform is designed to provide personalized learning experiences for students, supporting the construction of individualized learning paths that adapt in real time to student input and performance. The “Find out more” feature allows students to select a specific subsection of the course content they want to explore further, thus enabling them to delve deeper into topics of interest or dificulty in the form of a text conversation. This type of targeted just-in-time support has been shown to foster learner engagement and improve content retention [17]. The feature is showcased in Figure 2. When the feature is triggered, the AI tutor is invoked with a prompt requesting the LLM to provide an additional explanation regarding the topic of the current subsection. In order to ensure a relevant and accurate response, the LLM is provided with a contextual prompt including the subsection content and the most appropriate excerpts from relevant learning resources. This form of context-aware augmentation aligns with findings in [ 18, 19], which emphasize the importance of timely, context-sensitive AI feedback in educational systems.
Upon the initial explanation, students can continue conversing with the AI tutor. This feature, demonstrated in Figure 3, is particularly beneficial for supporting self-regulated learning, as students can request elaborations, ask clarifying questions, and explore related content in a conversational interface. It is important to note that the quality of the AI tutor’s answers depends on the underlying LLM used. The responses are not curated by the lecturer, and therefore, misinterpretations may occasionally occur. However, the use of tailored prompts with high-quality contextual information has been shown to reduce hallucinations in LLM outputs [16].
In addition to personalized content exploration, mISLec4EDU allows students to reinforce their knowledge through embedded practice quizzes, as presented in Figure 4. These quizzes are authored by the lecturer and can be inserted at any point within the learning content. The strategic placement of such assessments has been shown to enhance retention and meta-cognitive awareness [20]. Each question is accompanied by detailed feedback, automatically generated and aligned with the student’s selected response. When a student selects an incorrect answer, they are presented with a personalized explanation that targets misconceptions and reinforces correct reasoning, following evidence-based principles of formative assessment.
Upon completion of a quiz, students receive an overall feedback summary, which highlights their strengths and weaknesses and provides recommendations for revisiting specific topics. This tailored feedback loop supports mastery learning and encourages reflective learning behavior, which is closely associated with improved academic outcomes [21].
mISLec4EDU is designed not only with the learner in mind but also as a productivity and pedagogical
aid for educators, responding to the increasing demand for scalable authoring tools in digital learning
environments. One of its core features is the human-in-the-loop approach to assessment creation.
Rather than fully automating question generation, which risks producing irrelevant or misaligned
assessments, mISLec4EDU enables lecturers to co-create quiz content with AI support, preserving
pedagogical integrity while accelerating the design process [
The lecturer has the option to define the number of single-choice and multiple-choice questions for each practice or exam quiz. Additionally, they can specify which additional learning materials should be used as contextual input for the LLM when preparing suggested questions and answers. Based on this input, the LLM is invoked with a customized prompt that combines pedagogical objectives, curriculum alignment, and instructional design principles. The prompt includes relevant content excerpts and explicit instructions for generating well-structured questions and answers.
Prompt engineering in this context ensures alignment with educational goals and reduces the cognitive burden on teachers. As a result, as shown in Figure 6, the lecturer receives draft questions tailored to the course objectives. It is important to emphasize that these are merely suggested items. Therefore, the final responsibility for validation and adaptation remains with the educator, reinforcing academic control and safeguarding quality [22]. This approach ensures instructional rigor while significantly reducing the time and efort typically required in manual assessment design.
This paper presented mISLec4EDU, an AI-enabled platform that unifies asynchronous and personalized learning with pedagogical control and educator support. Unlike fragmented tools, it combines conversational tutoring, formative assessments, and AI-assisted content generation in a single environment. With supports both third-party and locally hosted large language models, ofers institutions flexibility in addressing infrastructure, cost, and data governance requirements. The platform emphasizes contextaware prompting, learner-specific feedback, and a human-in-the-loop workflow that enables educators to validate and adapt AI-generated materials to maintain instructional integrity.
While the system’s capabilities have been demonstrated through a detailed case study, we recognize that the current work does not include empirical validation. Claims about improved student engagement or reduced educator workload are grounded in design goals and system functionalities rather than experimental evidence. Future work will therefore focus on conducting systematic empirical studies to evaluate the platform’s educational impact. These will include comparative studies between traditional and AI-supported learning, usability testing with both students and educators, and longitudinal assessments of learning outcomes, retention, and engagement across diverse learner demographics.
In addition, we plan to engage more deeply with critical issues surrounding generative AI in education, including transparency of AI decisions, fairness across user groups, explainability of content, and the mitigation of hallucinations or misinformation. Enhancements to the platform will explore mechanisms to increase user trust and awareness of the limitations of AI-generated content while reinforcing the educator’s role in maintaining pedagogical rigor.
In conclusion, mISLec4EDU represents a step toward practical, flexible, and ethically grounded integration of generative AI in educational contexts. While not a substitute for traditional pedagogy, the platform supports the design of learner-centered experiences that can be scaled and adapted to institutional needs.
The authors acknowledge the financial support from the Slovenian Research and Innovation Agency (Research Core Funding No. P2-0057).
During the preparation of this work, the author(s) used Grammarly in order to: Grammar and spelling check. After using these tool(s)/service(s), the author(s) reviewed and edited the content as needed and take(s) full responsibility for the publication’s content. [4] Spanish Ministry of Education, Vet and Sports, Guide on the Use of Artificial Intelligence in Education, 2024. URL: https://code.intef.es/wp-content/uploads/2024/07/ Guidelines-on-the-use-of-AI-in-Education-INTEF_2024.pdf. [5] Commonwealth of Australia, Australian Framework for Generative Artificial Intelligence in Schools, 2023. URL: https://www.education.gov.au/schooling/resources/ australian-framework-generative-artificial-intelligence-ai-schools. [6] M. Alier, F. García-Peñalvo, J. D. Camba, Generative artificial intelligence in education: From deceptive to disruptive, International Journal of Interactive Multimedia and Artificial Intelligence 8 (2024) 5–14. [7] T. K. Chiu, B. L. Moorhouse, C. S. Chai, M. Ismailov, Teacher support and student motivation to learn with Artificial Intelligence (AI) based chatbot, Interactive Learning Environments 32 (2024) 3240–3256. [8] M. Jovanovic, M. Campbell, Generative artificial intelligence: Trends and prospects, Computer 55 (2022) 107–112. [9] K. Ackermans, M. Bakker, A.-M. van Loon, M. Kral, G. Camp, Young learners’ motivation, selfregulation and performance in personalized learning, Computers & Education 226 (2025) 105208. [10] D. Baidoo-Anu, L. O. Ansah, Education in the era of generative artificial intelligence (AI): Understanding the potential benefits of ChatGPT in promoting teaching and learning, Journal of AI 7 (2023) 52–62. [11] North Carolina Department of Public Instruction, North Carolina Generative AI Implementation Recommendations and Considerations for PK-13 Public Schools, 2024. URL: https://go.ncdpi.gov/ AI_Guidelines. [12] H. S. Rad, R. Alipour, A. Jafarpour, Using artificial intelligence to foster students’ writing feedback literacy, engagement, and outcome: A case of Wordtune application, Interactive Learning Environments 32 (2024) 5020–5040. [13] A. Andinia, I. N. Isnainiyah, Design of learning application using trivia method based on google assistant for vision impairment disability, in: 2020 international conference on informatics, multimedia, cyber and information system (ICIMCIS), IEEE, 2020, pp. 220–225. [14] N. Selwyn, On the limits of artificial intelligence (ai) in education, Nordisk tidsskrift for pedagogikk og kritikk 10 (2024) 3–14. [15] I. J. Akpan, Y. M. Kobara, J. Owolabi, A. A. Akpan, O. F. Ofodile, Conversational and generative artificial intelligence and human–chatbot interaction in education and research, International Transactions in Operational Research 32 (2025) 1251–1281. [16] X. Liu, J. Wang, J. Sun, X. Yuan, G. Dong, P. Di, W. Wang, D. Wang, Prompting frameworks for large language models: A survey, arXiv preprint arXiv:2311.12785 (2023). [17] J. L. Chiu, M. T. Chi, Supporting self-explanation in the classroom, Applying science of learning in education: Infusing psychological science into the curriculum (2014) 91–103. [18] R. Nakamoto, B. Flanagan, Y. Dai, T. Yamauchi, K. Takami, H. Ogata, Enhancing self-explanation learning through a real-time feedback system: An empirical evaluation study, Sustainability 15 (2023) 15577. [19] D. Gomes, A comprehensive study of advancements in intelligent tutoring systems through artificial intelligent education platforms, in: Improving Student Assessment With Emerging AI Tools, IGI Global Scientific Publishing, 2025, pp. 213–244. [20] H. L. Roediger III, J. D. Karpicke, Test-enhanced learning: Taking memory tests improves long-term retention, Psychological science 17 (2006) 249–255. [21] R. Morris, T. Perry, L. Wardle, Formative assessment and feedback for learning in higher education:
A systematic review, Review of Education 9 (2021) e3292. [22] G. Trajkovski, H. Hayes, Practical guide: Implementing ai-assisted assessment, in: AI-Assisted Assessment in Education: Transforming Assessment and Measuring Learning, Springer, 2025, pp. 381–416.