Past category matching,
into human context
An AI recommendation system that combines personal signals (birth date, MBTI) with multi-dimensional interests (level, role, category, capital, monetization) to suggest courses. Designed to surface 'what suits this person' rather than what category-matching can find.
AI · Recommendation · Education
Real screens.
Screens captured while the service was live. Select an image to enlarge it.
Recommend only by category, and users only see what they already know.
Category filtering keeps the user in familiar territory and turns recommendation into a short form of search. Real-life learners factor in their job, capital, monetization approach and other concrete conditions.
The operator was already seeing the data: category-only recommendation wasn't lifting next-stage course revenue.
Personal signals as small inputs, multi-dimensional interests as the main axes — combined so recommendations narrow inside each user.
Personal signals (name, birth date, MBTI) sit alongside interest signals (level, role/context, category, required capital, monetization) and connect the user's 'current position' to 'desired position.'
Beyond category sorting, the model factors in psychological traits and practical constraints. The focus was lifting the rate of users moving from one course to the next.
Stack.
Scikit-Learn · TensorFlow
Separate classification models for personal signals and interests.
Spring · MySQL
User signals and course metadata processed in one place.
React · AWS
Recommendation results UI and the operator admin.