Hi community,
I’ve developed an open-source platform called SmartDaaS that sits on top of DHIS2 and EMR CSV exports to generate patient-level treatment interruption risk scores, facility benchmarking, and programme intelligence — without changing existing workflows or requiring new data collection.
The platform accepts standard DHIS2 exports and outputs:
- Patient risk scores with SHAP explainability (why each patient is flagged)
- Risk-adjusted facility benchmarking
- Data quality grading before any analysis runs
- Executive reports formatted for PEPFAR and Global Fund reporting
The model was trained on 27,288 HIV patient records and validated across independent datasets from Kenya, Malawi, Rwanda, Tanzania, Uganda, and Zambia. Two preprints are available on medRxiv.
Live platform: https://smartdaas-hiv-validation.onrender.com
GitHub: GitHub - Kchinthala15/smartdaas-hiv-validation: AI-powered HIV programme intelligence — 27,288 Nigerian HIV Programme Records (Discovery Cohort) | Pilot-Ready | PEPFAR | Global Fund | Local Recalibration · GitHub
Preprint: https://doi.org/10.64898/2026.05.15.26353325
I’m looking for feedback from DHIS2 implementers and HIV programme data managers — particularly around variable mapping from DHIS2 exports and whether the approach addresses challenges you’re seeing in the field.
Happy to answer any questions.