Doctoral Research Project: The Maher AI Agent
Design, Implementation, and Evaluation of a Socratic Conversational Agent for the Development of Critical and Creative Thinking in Mathematics.
1. Context and Problem Statement
The integration of Artificial Intelligence in Education (AIEd) faces a major perverse effect: the widespread adoption by students of traditional chatbots or static assistants (Custom GPTs) that act as mere “answer dispensers.” This technological dependency fosters cognitive laziness and superficial memorization at the expense of deep cognitive skills. In mathematics, particularly during the transition period of the 6th grade (ages 12–13)—a critical juncture for moving toward abstraction and formal reasoning—digital tools frequently fail to stimulate the higher-order executive functions required to solve complex problems.
2. Thesis Statement
This research posits that integrating an intelligent conversational agent (AI Agent) acting as a Socratic “thinking coach” optimizes learning through individualized and dynamic pedagogical scaffolding. By refusing to provide raw solutions and finely adjusting its level of guidance based on the learner’s micro-errors, this agent significantly promotes the development of critical (Analysis, Interpretation, Inference) and creative (Fluency, Flexibility, Originality) thinking skills, far beyond the algorithmic application of rules.
3. Technological and Pedagogical Innovation: The Maher Agent
To validate this thesis, the project involves developing Maher (“the skillful”), an intelligent agent featuring a dual breakthrough:
- A Hybrid Deterministic Architecture (DAG Engine): Unlike free-generation LLMs prone to mathematical hallucinations, Maher relies on a Directed Acyclic Graph (DAG) to drive rigorous decision trees and standardized, three-level Socratic scaffolding (Focusing, Deconstructing, Metacognitive Anchoring).
- True Genericity and Autonomy: Maher integrates a multimodal ingestion pipeline (Visual-to-Code) capable of dynamically extracting the structure of any textbook. It thus adapts in real time to the official curriculum, providing targeted interventions following the initial instructional phase led by the teacher.
┌────────────────────────────────────────────────────────┐
│ Ingestion Pipeline (VPS) │
│ Visual Analysis and Symbolic Math of Textbooks │
└───────────────────────────┬────────────────────────────┘
▼
┌────────────────────────────────────────────────────────┐
│ DAG Engine (Deterministic Logic) │
│ Dynamic Adjustment of Socratic Scaffolding Level │
└───────────────────────────┬────────────────────────────┘
▼
┌────────────────────────────────────────────────────────┐
│ Tablet Application (Flutter Client) │
│ Tactile Interactions, Offline Mode & xAPI Traces │
└────────────────────────────────────────────────────────┘
4. Experimental Design and Field Realism
The empirical evaluation of Maher is designed to accommodate the logistical and regulatory constraints of real-world school environments:
- Hardware Scarcity Management: To address classroom realities (modeled here on a ratio of 40 students to 10 tablets), the application integrates a rapid rotation protocol using secure authentication (QR Code/NFC badges). Student sessions are rigorously partitioned on the same device.
- Network Resilience and GDPR Compliance: The application operates autonomously in offline mode, utilizing a local encrypted database (AES-256 via SQLCipher) to protect minors’ sensitive data. Data is synchronized asynchronously to a dedicated VPS server (KVM 2).
- Standardized Quantitative Methodology (xAPI): Every student action, hesitation, error, or success is captured as standardized Actor-Verb-Object micro-data (xAPI). These traces feed automated statistical analysis routines to precisely calculate the learners’ fading index (the progressive reduction of AI assistance) and overall Learning Gain.