MasteryMap AI — Explainable Adaptive Learning
The education problem
Students often spend equal time on every topic even when their knowledge gaps are very different. Many recommendation systems also hide why they selected the next activity. MasteryMap AI addresses both problems: it continuously estimates concept mastery, chooses the next question based on learning value, and explains the recommendation to the learner.
Educational impact
MasteryMap tracks mastery across Algorithms, Networking, Databases, and Probability. After each answer it updates the learner’s estimated knowledge and immediately adapts the next activity. The goal is to spend practice time where it can create the most learning value instead of following a fixed sequence.
The interface exposes the reason for every recommendation so the learner can understand why a topic was selected. That makes adaptation itself part of the learning experience rather than an invisible ranking system.
AI / ML is the core, not an add-on
The adaptive engine uses Bayesian Knowledge Tracing (BKT) for per-concept mastery updates. It models learning with explicit parameters for prior mastery, learning, guessing, and slipping. Question selection then combines several signals:
- current mastery / learning need;
- model uncertainty;
- predicted probability of success;
- difficulty fit;
- spacing urgency;
- novelty / previously used items.
This means the ML model directly changes what the learner studies next. The recommendation is deterministic and explainable rather than a decorative chatbot layer.
Explainable decisions
For each selected question, MasteryMap shows why the model chose it, including mastery state, uncertainty, spacing and predicted-success information. The student can inspect the system’s reasoning instead of being asked to trust a black box.
Technical execution
The Prom Fall Classic build runs fully in the browser using JavaScript, HTML and CSS. Learning state is stored locally, so the demo requires no account, external API key, or uploaded learner data. The project includes automated model tests covering mastery updates and recommendation behavior.
The submitted experience includes:
- a complete adaptive question loop;
- per-concept and overall mastery tracking;
- predicted-success feedback;
- explainable next-question selection;
- responsive mobile/desktop UI;
- 12-question learning set across four technical subjects;
- local progress persistence;
- deterministic model tests.
Why privacy matters in education
The current demo keeps progress in the browser and does not upload learner answers or profiles. This keeps the prototype simple to inspect and demonstrates that meaningful adaptive learning does not require collecting unnecessary personal data.
Challenges and learning
The hardest design problem was balancing adaptation with transparency. A more complex model is not automatically a better educational product if learners and teachers cannot understand its decisions. We therefore designed the ranking system so its major factors are inspectable and explainable.
We also learned that adaptive learning becomes more useful when the system communicates uncertainty. The model is not claiming to “know” the student; it maintains an estimate that changes as new evidence arrives.
Pitch and demo
The demo focuses on the complete learning loop: answer a question, observe the mastery estimate change, inspect the model’s reasoning, and see the next recommendation adapt. This directly demonstrates both the educational value and the ML mechanism rather than only showing static screens.
Built during the hackathon
The core adaptive-learning application logic and the submitted Prom Fall Classic experience were created during the hackathon build window.
What's next
Next steps include larger teacher-authored question banks, richer spaced-repetition scheduling, curriculum import, teacher/classroom analytics, calibration studies with real learning data, and optional local or hosted language-model support for generating explanations and practice items while preserving the transparent mastery model.
Source code
Built With
- bayesian-knowledge-tracing
- css3
- html5
- javascript
- localstorage
- machine-learning
Log in or sign up for Devpost to join the conversation.