The question.
Learning recommendations are weak when the system cannot explain the prerequisite a learner is missing or distinguish fluency from understanding.
The expanded Pathway concept combines explicit curriculum structure with an AI tutor grounded in approved material. It avoids treating engagement as proof of learning.
The proposed approach.
- Represent concepts, prerequisites, and learning objectives as a versioned graph
- Ground tutor explanations in the course material
- Use formative questions to estimate understanding with uncertainty
- Give instructors control over content, interventions, and evaluation
Target architecture.
Course materialCurriculum graphLearning evidenceTutor / recommendationsFormative assessmentInstructor view
The difficult parts.
- Noisy mastery signals and sparse learner histories
- Unsupported explanations or over-helpful answers
- Balancing personalization with instructor intent
How to evaluate it.
- Tutor grounding and instructional quality on reviewed tasks
- Understanding measured through suitable assessments
- Recommendation usefulness across different starting knowledge
Planned deliverables.
- A curriculum graph and learning-event model
- A grounded tutoring workflow
- An instructor-centered evaluation plan
Design principles.
Adapt to evidence of understanding, not just clicks.
A tutor should help a learner reason, rather than simply finish the task.