A Provider-Independent LLM Architecture for Adaptive Oracle SQL Tutoring: A Design Study

Abstract

Adaptive learning environments and large language models (LLMs) are increasingly combined, yet most LLM-based tutoring work addresses feedback generation in isolation rather than its integration into a controlled, auditable learning loop. This paper presents a design study of a provider-independent LLM architecture for adaptive Oracle SQL tutoring. We report the system at the design stage—specifying the underlying adaptive learning platform, a provider-independent service layer that abstracts interchangeable model back-ends behind a single auditable interface, the SQL-specific extension (a reference-based evaluator demonstrated as a proof-of-concept on an in-memory SQLite fixture of the fixed Oracle HR schema; Oracle-specific execution, the live sandbox, and course-wide evaluation remain planned), and a concrete validation pathway—without claiming empirical effectiveness. The design pairs cognitive adaptivity, realised as mastery-based task selection under configurable parameters, with motivational design grounded in Keller’s ARCS model. Following a design-based research orientation, the adaptation logic is expressed as testable design conjectures. We contribute (a) a modular, provider-independent architecture for LLM-supported SQL learning; (b) a formalised adaptation logic linking learner states, adaptive responses, and conjectures; (c) an illustrative prompt template with a three-level quality-control workflow for AI-generated feedback; and (d) a proposed estimation-oriented feasibility pilot protocol—a validation pathway rather than an empirical contribution of this article—to be preregistered before data collection, with up to 50 undergraduate participants. Implemented capabilities, planned SQL-specific functions, and hypothesised outcomes are kept explicitly distinct throughout.