nspiration Thousands of university students fail or withdraw every year — not from lack of ability, but because support arrives too late. We wanted to flip that: predict the crisis before it happens.
What it does AcademicPulse identifies at-risk students up to 10 weeks before failure and automatically generates a personalised, Claude-powered intervention plan for each one — giving advisors actionable guidance, not just a warning.
How we built it Hybrid deep learning architecture combining a Bidirectional LSTM (for weekly behavioural sequences) with a static feature branch (demographics, assessments). XGBoost runs in parallel as an interpretable baseline. SHAP explains individual predictions. Claude generates clinician-quality intervention plans. Built on the OULAD dataset — 32,000+ students.
Challenges we ran into Class imbalance between pass and fail students. Mapping SHAP values back to human-readable feature names. Generating intervention plans that feel personal, not algorithmic.
Accomplishments that we're proud of End-to-end pipeline from raw data to AI-generated advisor reports. A fully interactive dashboard with zero backend required.
What we learned Prediction without explanation is useless. The Claude integration transformed a model output into something an actual advisor could act on.
What's next for ACADEMIC PULSE Real-time integration with university LMS platforms. Fairness auditing across demographic groups. Pilot deployment with a live institution.
Log in or sign up for Devpost to join the conversation.