Open Category Submission

833 – A full-fledged Intelligence Tutoring System (Comprehensive Framework)

This project proposes a very innovative Intelligent Tutoring System framework that integrates both affective and cognitive responses to recommend personalized learning paths and dynamically generated worksheets. Unlike most ITS models, which depend heavily on static repositories or pre-defined rules, this system employs artificial intelligence, namely natural language processing and affective computing, to glean not only what a student knows but also what he or she feels during the process of learning.

The major innovation lies in the response engine that operates in dual mode. While being used, the system continuously monitors students' interactions (e.g., text input, time taken on task, and correctness), while also applying emotion detection techniques through sentiment analysis along with behavioral cues to identify feelings of frustration, boredom, or engagement in the student. All these are pooled into a recommendation system that suggests interventions that may include switching question types, giving scaffolded hints, or encouraging motivational feedback.

On the technical side, the system integrates LLMs with a cognitive diagnostic model (CDM) that is informed by the Knowledge-Learning-Instruction (KLI) framework. This allows for real-time generation of personalized worksheets that address diagnosed misconceptions, Bloom's Taxonomy progress levels, and detected affective states. Worksheets are automatically customized for content, difficulty, and mode of delivery, creating an individualized, emotionally aware learning experience.

From a design perspective, the intuitive teacher dashboard, which visualizes student cognitive and affective states, gives teachers insight into class perspectives as well as individual needs. Interventions can be made by teachers and algorithmic behavior adjusted, or simply collaborative elements incorporated without ever feeling displaced.

The potential impact of this project is huge. This aligns learning materials with both cognitive readiness and more immediate affective state, thus increasing student motivation and retention. It further supports inclusive differentiated instruction at scale. In advancing an affect-aware adaptive learning environment, this system, therefore, propels ITS closer toward reaping the benefits of tutoring on a classroom or institutional scale.

My project is motivated by the increasing gap between the demands of learners in intricate, real-world educational contexts and the Intelligent Tutoring Systems (ITS) that are currently in use. The majority of systems are constrained by either strict exercise repositories or highly scaffolded problem-solving pathways, both of which restrict learner agency and higher-order thinking, according to my experience creating personalized learning tools and a critical analysis of recent ITS literature. These models perform poorly in open-ended, interdisciplinary, or affect-driven contexts, despite their success in structured domains.

The idea of utilizing AI, particularly large language models, to transition from static content delivery to dynamic, real-time adaptation particularly appealed to me. My hypothesis is that an ITS enhanced by affective computing and cognitive diagnostic capabilities can identify and address a learner's misconceptions, emotional state, and progression through Bloom's Taxonomy, with the Knowledge-Learning-Instruction (KLI) framework (Koedinger, 2012) serving as a theoretical foundation. This preserves the coherence of instruction while enabling more autonomy and more effective feedback.

As evidenced by systems where students design exercises, integrating student-generated content also presents an opportunity for deeper engagement; however, current implementations lack intelligent scaffolding and feedback mechanisms. Both independent and group learning can be supported by an ITS that fosters learner-driven inquiry and AI-driven feedback, all while strengthening the role of the instructor.

I think this approach will work because it blends the flexibility of generative AI and structured ITS approaches with the cognitive underpinnings of educational psychology. It also bridges significant gaps in assessment, personalization, and teacher integration, which enables it to be tailored to a range of learning environments.

The self-same encouraged solution will be implemented on a stack of proven and scalable technologies. Integration at the core of it all is the large language models (LLMs) through APIs (say OpenAI or open-source models hosted on AWS/GCP) for natural language understanding, personalized feedback and generative worksheet creation. Emotion recognition algorithms will drive the affective computing based on sentiment analysis from text and optionally, facial or voice analysis (e.g., using open-source toolkits such as OpenFace or Microsoft Azure Emotion API).

For early development, initial build will be hosted on AWS with a modular backend (Node.js or Python Flask) and a front-end developed using React for an interactive dashboard. User management and real-time data sync features can be achieved using Firebase or Supabase. Build resources will comprise one AI engineer, one full-stack developer, and an education consultant to align pedagogical goals with user functionality.

Core Product Features include:

Real time detection of affective state and cognitive performance.

Adaptive recommendation of next learning task worksheets.

Intelligent Feedback based on concept-level diagnostics.

Teacher dashboard to visualize learner data and possible intervention.

Student interface supporting task input, reflection, and review.

We will run a defined pilot scheme across tuition centres and secondary schools already experimenting with some form of AI teaching to validate the market demand. Surveys and focus groups will measure desirability and usability. At the same time, online performance metrics will assess the system impact in terms of: (i) learning gain (pre/post assessment), (ii) time-on-task improvement, (iii) engagement scores (via sentiment tracking), and (iv) recommendation accuracy (informed by learning outcomes).

User experience will go through iterations of change through UI testing and feedback from students and teachers. In this way, we design together with teachers focused on classroom adoption, complementing instead of hindering pedagogical roles so that functional, sustainable deployment happens in genuine learning environments.

N/A

This project is a revolutionary step towards the progress of ITS by going an additional mile beyond heights for an emotionally and cognitively adaptable learning model in Intelligent Tutoring Systems. Most ITS are actually about diagnosing gaps and prescribing remedial exercises without consideration for the students' emotional states and with really good evidence that affect influences the learning outcome.

Our system is of a unique convergence of affective computing with cognitive diagnostics into a real-time AI recommendation engine. The adaptivity is therefore two-tier where the system identifies confusion, dis-engagement, or even anxiety in the students, and the interventions are contextually appropriate: changing the task's difficulty, providing encouragement, or switching formats. This would be a paradigm shift in creating on-the-faces personalized concept-targeted worksheets using generative AI instead of static content banks.

Another very interesting payoff for creativity is the transition from the teacher imposed model to the learner-initiated pathway where students can interact with open-ended tools that help to support task generation and exploration while the system maintains pedagogical coherence through guided feedback. The new technology is not only automation but augmenting human-like intuition with knowledge in the use of Knowledge-Learning-Instruction (KLI) as cognitive scaffold.

In fact, the teacher-facing dashboard is a creative integration into the classroom. Rather than displacing teachers, it provides them with visualized learner data to enable timely intervention and collaborative planning. This human-AI partnership fosters confidence and continued sustainability.

All these further improve the system's efficiency by customizing support to the cognitive profile and emotional readiness of the learners on an individual basis—enhancing engagement, deepening understanding, and fast-tracking learning pathways. It tackles scalability, personalization, and equity in critical ways through this human-level adaptivity at scale, establishing a new standard for emotionally intelligent learning technologies.

This modular cloud-native architecture is thus made scalable from the get-go using containerized services such as Docker or Kubernetes and scalable infrastructure such as AWS Lambda or Google Cloud Functions which will self-adjust to the changing user load. It uses API-based LLMs (e.g., OpenAI, Cohere) whereby they are going to be elastically scaled for compute-intensive tasks, such as feedback generation. Plus, it will keep down the number of API calls and hence the cost by caching responses commonly accessed.

Emotion and performance analytics will be batch processed through data pipelines to limit real-time computation. Use offline first features like local storage and edge caching for deployment in poorly connected countries. The modular plugin architecture of the system would also allow for quickly adding new subjects, formats of exercise, or school-specific features without having to rewrite the entire system.

Interleaving of bottlenecks like LLM inference latency or increased data load will be solved through asynchronous task queues, load balancers, and hybrid modeling options (for example, fine-tuned lightweight models for high-frequency tasks). As user demand grows, it is intended to further support integration into these school LMS platforms through partnerships with institutional IT departments and EdTech vendors.

In a pursuit of environmental sustainability, the platform seeks to optimize for computing efficiency by dynamically switching from heavy to light inference models. Preference will be granted to cloud providers with an equal commitment to being carbon neutral (e.g., AWS or GCP). The event-driven architecture will minimize idle compute resources.

The system will entail learner goal-setting, reflection logs, and gamified mastery tracking for long-term user engagement, and, on the other hand, actionable insights towards instructional planning will be offered by means of the teacher dashboard. Feedback loops with users—surveys embedded within the product and A/B testing—will ensure constant alignment of all product evolution with user needs and curriculum changes.

The solution is calibrated for sustainability in heterogeneous and ever-changing education settings, combining technical scalability, ecological responsibility, and adaptive design.

This project seeks to rectify systemic educational inequalities by developing a tutoring system that is sensitive to the needs of students, considering not only their cognitive abilities but also their emotional character. Thus, the entire learning experience will be rendered more human-centered and equitable. Its major beneficiaries are students originating from underserved communities, who for various reasons—mainly socio-economic and geographic—do not have access to one-on-one or personalized tutoring or support.

With an AI-based low-cost solution, which mimics key features of personalized instruction, this system allows these learners to get individualized feedback, scaffolding, and motivation no matter how large the class is or unavailable the instructor might be. The inclusion of affective computing is most useful for those who silently struggle; that is, the system can detect disengagement, frustration, or confusion and provide appropriate supportive intervention.

The platform promotes even more inclusion and accessibility by being usable across devices—particularly low-end mobile phones—and supporting several languages. Rather, the companion dashboard includes useful insights into students' learning gaps and emotional trends, empowering teachers to respond more quickly and empathically in their instruction.

To ensure circumstances aim at wider societal goals, the roadmap for the development of this system includes provisions for the use of culturally inclusive content examples, bias detection within AI feedback, and accommodation for neurodivergent learners.

The project's social impact will include both qualitative and quantitative metrics that will measure:

Changes in the learning outcomes of disadvantaged groups of beneficiaries (for example, standardized test scores and learning gains)

User retention and emotional engagement metrics

Accessibility and utilization metrics across diverse demographics

Teacher-reported improvement of inclusive and engaging classroom environments

The feedback will continue with a series of co-design sessions through which educators, students, and community stakeholders can engage with the design and co-create programs ensuring real-time development according to real-world educational challenges. Thus, this project aspires to democratize emotional intelligence-high-quality instruction and help bridge an equity gap that presents itself in global education.

Personal Information Collection Statement (PICS):
1. The personal data collected in this form will be used for activity-organizing, record keeping and reporting only. The collected personal data will be purged within 6 years after the event.
2. Please note that it is obligatory to provide the personal data required.
3. Your personal data collected will be kept by the LTTC and will not be transferred to outside parties.
4. You have the right to request access to and correction of information held by us about you. If you wish to access or correct your personal data, please contact our staff at lttc@eduhk.hk.
5. The University’s Privacy Policy Statement can be access at https://www.eduhk.hk/en/privacy-policy.

• I have read and agree to the competition rules and privacy policy.