Adaptive Learning Engine with Real-Time Analytics
Engineered a personalized spaced-repetition engine powered by Bayesian Knowledge Tracing, processing 2M+ daily learning events to raise course completion by 41%.
2M+ Daily Events+41% Completion RateReal-Time Dashboards
Key Takeaways
- Dynamic Bayesian Knowledge Tracing (BKT) tailored content difficulty to individual student mastery curves.
- High-throughput Kafka and ClickHouse pipelines processed 2M+ daily interactions with sub-second dashboard updates.
- Student Net Promoter Score (NPS) rose from +12 to +47, with average time-to-mastery compressed by 33%.
The Challenge
An online education platform serving 300,000 students delivered static curriculum pathways. Course completion rates hovered at 34%, as beginner students were overwhelmed by rapid difficulty spikes while advanced learners abandoned early lessons due to repetitive content.
Architecture & Technical Approach
- Bayesian Knowledge Engine: Maintains a real-time probability vector of skill mastery for each student across 1,400 curriculum concepts.
- Event Streaming Architecture: Apache Kafka streams student interactions (quiz answers, hint clicks, video pauses) into a Go worker pool.
- Columnar Analytics: ClickHouse aggregates student metrics, powering real-time instructor dashboards with sub-50ms query latency.
Quantitative Benchmarks & Results
| Educational Metric | Static Curriculum | Adaptive BKT Engine | Outcome |
|---|---|---|---|
| Course Completion Rate | 34.0% | 48.1% | +41.5% Completion Lift |
| Average Concept Mastery Time | 6.2 Weeks | 4.1 Weeks | 33.8% Faster Mastery |
| Student Net Promoter Score (NPS) | +12 | +47 | +35 Point Lift |
| Analytics Data Freshness | 5 Days (Batch CSV) | < 1 Second (Streaming) | Real-Time Feedback |
Production Reliability & Lessons Learned
Structuring spaced-repetition reminders based on individual forgetting curve decay rates outperformed static calendar schedules by 2.4x in 90-day retention tests.