An Intelligent LLM-Based Personalized Tutoring Framework for Adaptive Education

1. Problem

Large language models are entering education as chat companions, yet most deployments are not tutors: they lack an explicit learner model, a pedagogical policy for when to hint versus tell, and a teacher-facing oversight path. Classical intelligent tutoring systems (ITS) offered adaptivity with heavy authoring cost and brittle dialogue. LLM chat reverses the trade-off — fluent language, weak control — and can hallucinate on technical material. A doctoral contribution is needed that treats personalization as a closed loop suitable for adaptive education, not as open-ended conversation over course notes.

2. Aim and research questions

Aim: Design, implement, and evaluate an intelligent LLM-based personalized tutoring framework for adaptive education that diagnoses learner state, selects pedagogical actions, generates constrained dialogue, updates the learner model, and remains overridable by a human teacher.

1. How should an LLM tutor represent mastery and misconceptions so adaptation is inspectable?
2. Which pedagogical policy (hint, example, quiz, remediate, escalate) improves learning gains versus static LLM chat?
3. How can teachers oversee and correct the tutor without rewriting every turn?

3. Contribution

A reusable tutoring framework (architecture + policy + oversight) and an evaluation protocol showing learning gains and bounded over-reliance on one curriculum testbed. Optional domain grounding for technical syllabi (e.g. IT infrastructure or aviation-related training modules). Artefacts: open logging schema, scenario/assessment pack, and lab working paper.

4. Method (overview)

Year 1 locks literature, syllabus, and ethics. Year 2 builds the framework prototype and pilots. Year 3 runs a controlled comparison: framework tutor vs vanilla LLM chat vs materials-only, with pre/post tests, delayed retention where feasible, interaction logs, and teacher audit time. Year 4 completes the thesis and a venue paper (AIED, Learning@Scale, or related).

5. Testbed and candidate context

The NEXORA-iQ Learning Hub (AI paths and system-administrator programmes) provides an immediate educational setting and live-class community. Professional content from aviation and IT infrastructure backgrounds may ground technical tutoring chapters without turning the thesis into a product. One primary syllabus is locked before the main study.

6. What this thesis is not

It is not a commercial ed-tech product, not a foundation-model training run, not a prompt library, and not a satisfaction-only chatbot evaluation. Fluency of the LLM is assumed; control, personalization, and learning outcomes are the scientific objects.

7. Near-term plan (lab checkpoints)

8. Indicative references

1. VanLehn, K. — the behaviour of tutoring systems / ITS loop.
2. Knowledge tracing and learner modelling literature for mastery over knowledge components.
3. Scaffolding and cognitive apprenticeship — fading support in instructional dialogue.
4. Recent AIED / CHI / Learning@Scale studies of LLM tutors with learning-outcome measures.
5. Human-in-the-loop teaching and teacher oversight of AI classroom tools.

NEXORA-iQ community lab working paper — educational research track, not a commercial consultancy proposal. Topic brief: /research/llm-personalized-tutoring/