Research-backed design
ArogyaGrid combines lessons from LMIC medicine stock-out literature, digital LMIS adoption, intermittent-demand forecasting, graph optimization and Indian public-health systems.
Finding 1
Research on LMIC medicine stock-outs shows last-mile gaps often come from upstream forecasting, replenishment and visibility failures, not only local negligence.
Product decision:Facility scoring uses attribution: local, upstream, systemic or uncontrollable.
Finding 2
Digital LMIS value depends on fitting existing workflows. ArogyaGrid therefore uses voice, IVR, register photos and existing system imports.
Product decision:Passive capture is mandatory; staff form-filling is avoided.
Finding 3
Facility demand is sparse, seasonal, intermittent and stock-out-censored. One LSTM model is not credible for all items.
Product decision:Use quantile GBM, seasonal baselines and Croston/TSB methods by data quality.
Finding 4
Alerts must become feasible transfer orders that respect buffer stock, expiry, routes, cold-chain and emergency reserves.
Product decision:Model district facilities as a graph and solve transfer recommendations as min-cost flow.
Finding 5
DVDMS/e-Aushadhi, HMIS, ABDM/HFR and IPHS should be leveraged rather than replaced.
Product decision:ArogyaGrid is a decision-support layer over those systems.
Finding 6
This use case does not need patient names, ABHA IDs or longitudinal records.
Product decision:Store aggregate operational signals; delete raw images after extraction.
Source set used