I built SmartDaaS because many HIV programmes only identify treatment interruption after patients have already disengaged from care. I wanted to see whether routine programme data already being collected could be transformed into practical decision support without requiring new infrastructure or workflow changes.
SmartDaaS was built as a solo, bootstrapped project and trained on real-world HIV programme data with external validation across multiple countries.
Happy to answer questions about the methodology, explainability (SHAP), validation approach, or implementation design.
SmartDaaS predicts HIV treatment interruption risk from routine DHIS2 and EMR exports — no new data collection, no workflow changes.
Built for PEPFAR implementing partners across sub-Saharan Africa.
- Trained on 27,288 real-world HIV patient records
- Temporal AUC 0.772 on held-out future patients
- SHAP explainability for every risk score
- No data stored — session-only processing
- Validated across 192,732 records in 6 countries
- MIT licensed
