EducAgent -- Graph-Grounded Agentic Pedagogy for Hallucination-Prone Domains

Understanding causality is essential to science.

From clinical trials to epidemiology to AI, getting causality right changes outcomes. When researchers get causality

outcomes. When researchers get causality wrong, they draw wrong conclusions from correct data. Yet, teaching it

correct data. Yet, teaching it effectively remains an unsolved challenge.

Textbooks are sequential. They can't

adapt to what you already know.

AI tutors hallucinate on technical content and students can't tell. Current

online learning platforms are one-sizefits-all. They can't support the

one-sizefits-all. They can't support the interconnected reasoning that causality demands.

For the first time, three capabilities have matured simultaneously.

Large language models that can reason.

Knowledge graphs that encode domain structure. Multi- aent frameworks that

structure. Multi- aent frameworks that coordinate specialist roles. Education

sits at the intersection of three innovations that haven't been combined before. An expert knowledge graph as the

before. An expert knowledge graph as the ground truth, a learner profile that captures your background and multi-agent teaching grounded in the knowledge graph with materials that adapt to who you

are.

The knowledge graph encodes a comprehensive network of causal inference concepts and their relationships, including which concepts students most commonly confuse. Every AI

response is anchored to this graph.

Nothing is generated freely. The system

knows who you are. A clinician and a computer scientist learning causal inference need completely different analogies.

Educent combines adaptive course generation with active Socratic teaching. Content adapts to who you are.

teaching. Content adapts to who you are.

Personalized modules, inline quizzes, and examples drawn from your domain.

Three agents drive the Socratic loop.

The tutor explains, the teachback agent challenges with deliberate misconceptions, and the critic evaluates.

The Prodigge effect, correcting an agent's mistakes, cements your own understanding.

Today, learners can explore the full causal inference knowledge space.

Navigate any concept, see its neighbors, read graph grounded explanations.

The causal knowledge graph is built.

Learner profiling is underway. Next,

adaptive curriculum and Socratic teaching to personalize every learner's path through the knowledge space.

Developed at the University of Edinburg with Chihub UK as our first learning community. Building the infrastructure

community. Building the infrastructure for expert level personalized scientific education at scale.