Guide
How Knowledge Graphs Transform Curriculum Design
Traditional curriculum organizes content by lessons and chapters. But lessons are just delivery mechanisms—what students actually learn are concepts. Knowledge graphs reorganize education around these fundamental building blocks, revealing prerequisite relationships and creating cross-curriculum connections that linear syllabi cannot.
When you import a curriculum into a knowledge graph, AI identifies the underlying concepts in each lesson and learning objective. It then discovers connections: which concepts must be understood before others, which appear across different subjects, and which serve as foundational building blocks for advanced topics.
The result is a unified map of knowledge that enables truly personalized learning paths. Instead of every student following the same sequence, paths adapt based on individual mastery. When a student struggles, the graph traces back to find prerequisite gaps—the root cause, not just the symptom.
This approach also enables intelligent concept reuse. "Addition" taught in first grade math connects naturally to "Addition of Fractions" in fourth grade and "Addition of Vectors" in physics. The system recognizes these connections automatically, building a coherent understanding across subjects and years.
Guide
AI-Powered Assessments: From Intent to Insight
Creating quality assessments is time-consuming. Writing questions, ensuring appropriate difficulty, covering the right Bloom's levels, avoiding duplicates—teachers spend hours on work that could be accelerated with intelligent assistance. AI-powered assessment generation changes this equation.
The process starts with teacher intent expressed in natural language: "Create a 10-question formative quiz on fractions for my Grade 4 class, focusing on comparing fractions with different denominators." The AI parses this intent and develops an assessment strategy—determining question count, Bloom's level distribution, difficulty calibration, and which concepts to target.
Questions are sourced intelligently. The system first searches an existing question bank for validated, well-tested items that match the criteria. It checks freshness to avoid questions recently used with the same students. When no suitable question exists, it generates new ones with appropriate distractors and rationale.
Critically, teachers remain in control. Every generated question is presented for review with alternatives. Teachers can approve, swap, edit, or regenerate any item. Nothing reaches students without explicit approval. The AI amplifies teacher expertise—it doesn't replace teacher judgment.
Over time, the system learns from teacher feedback. Questions that are frequently swapped are improved. Approval patterns shape future suggestions. The question bank grows with validated, classroom-tested items organized by concept, difficulty, and cognitive level.
Guide
Understanding Student Mastery: Beyond Grades
A 75% test score tells you a student got three-quarters of the answers right. It doesn't tell you which concepts they understand, which they're guessing on, or why they got specific questions wrong. Traditional percentage scores hide more than they reveal.
Bayesian Knowledge Tracing (BKT) provides a more sophisticated model of student understanding. Instead of simple percentages, BKT estimates the probability that a student has truly mastered a concept based on their response history—accounting for guessing, slipping, and learning over time.
The model considers four key parameters: the probability of learning a concept after practice (learn rate), the probability of a correct guess without understanding (guess rate), the probability of an error despite true knowledge (slip rate), and the initial probability of prior knowledge. These parameters are calibrated per concept based on actual student data.
This approach enables genuine mastery tracking. Students progress from Novice (0-25%) through Developing (25-50%) and Proficient (50-75%) to Master (75-100%) based on sustained demonstrated understanding—not lucky guesses. The system also indicates confidence levels: high confidence with 25+ data points, low confidence with sparse data.
Perhaps most importantly, this model enables misconception detection. When students answer incorrectly, their response patterns are analyzed to identify common misunderstandings. Instead of just marking answers wrong, the system identifies the specific conceptual error and recommends targeted remediation.
