Teachy-TECH: An Explainable GenAI Framework for Adaptive Learning
The integration of generative artificial intelligence (GenAI) into adaptive learning environments presents challenges related to instructional grounding, learner proficiency estimation, prerequisite-aware adaptation and the explainability of pedagogical decisions. This paper introduces Teachy-TECH, a graph-informed adaptive learning framework that integrates two complementary computational planes connected through a shared Learning Graph: (a) a generative plane combining Retrieval-Augmented Generation (RAG), graph-derived constraints and instructor-governed content authoring; (b) a decision plane incorporating a Graph-informed Elo (GE) proficiency model, misconception diagnosis and two-level adaptive algorithms. Macro-adaptation recommends learning concepts and activities based on prerequisite readiness and estimated proficiency, while micro-adaptation selects assessment items according to predicted success, cognitive complexity and learner interaction history. Functional validation is conducted using 100 synthetic learners interacting with an SQL curriculum, generating 2,900 assessment responses, complemented by an offline policy evaluation involving 300 synthetic learners. Compared with the initial proficiency estimator, the proposed framework achieved lower RMSE (0.425 versus 0.513) and expected calibration error (0.038 versus 0.178). Graph-informed adaptation satisfies the tested recommendation and assessment-selection constraints, while all 980 adaptive item selections and 13 diagnostic-probe decisions are reproducible from the event log. The results establish the functional consistency and technical reproducibility of the implemented framework.
Authors
- Simona‐Vasilica Oprea (ORCID: https://orcid.org/0000-0002-9005-5181)
- Adela Bârã (ORCID: https://orcid.org/0000-0002-0961-352X)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-12
- DOI
- https://doi.org/10.5281/zenodo.24155783
- Primary Topic
- Intelligent Tutoring Systems and Adaptive Learning
- Type
- article
- Field-Weighted Citation Impact
- 0.00