Answers are grounded in this paper and cite their sources.
Functional evaluation demonstrates mentor-mode viability and generative limitations in an educational AI agent, highlighting technical feasibility and constraints of adaptive tutoring.
This paper presents the development and functional evaluation of OiQadam AI v0.2, an educational AI agent for school mathematics designed to preserve the learner's own reasoning rather than immediately provide a complete solution. Its pedagogical behaviour is configured through Personality, Rules, and Skills and is supported by a context-adaptive hint ladder together with Cognitive Pause, Teach Back, and Transfer Check mechanisms. The prototype was implemented in the ASI:One / Agentverse ecosystem and examined through 11 functional, subject-specific, and stress- test scenarios. Testing confirmed several mentor -mode functions while also exposing limitations of the generative approach, including imprecise error diagnosis, role-override vulnerability, ambiguous mathematical notation, and dependence of document retrieval on source format. Detected failures were retained as research evidence and used to revise the agent configuration. The study demonstrates the technical feasibility of a pedagogically constrained AI agent but does not claim a proven improvement in learner independence without a separate controlled study with students.
No takes yet. Share an insight, caveat, or question.
Dauletbayeva et al. (2026) studied this question.