Diverse EdTech Recommendations: Maximal Marginal Relevance

Personalized-learning platforms don't just rank resources by relevance — a list of the ten most similar practice problems to a student's last question is usually ten near-duplicates of the same subtopic. This simulator renders a 2D embedding space of learning resources clustered by topic, places a "student query" vector representing their current need, and runs a real Maximal Marginal Relevance selection loop live: each step picks the resource that maximizes a weighted combination of relevance to the query and dissimilarity to everything already on the list. Drag the λ slider to watch the recommended list swing between a tight, relevant-but-repetitive set and a broad, diverse-but-less-targeted one, and compare it live against a naive top-relevance-only recommender running on the same catalog and query.