Code for Communities · Smart Health

Voice-first AI command centre for every PHC and CHC.

ArogyaGrid turns WhatsApp voice notes, register photos, IVR calls and existing e-Aushadhi/HMIS signals into stock-out warnings, demand forecasts, redistribution orders and fair facility intervention flags.

Run the AI intake flow Open district dashboard

Hindi + Marathi Rs 6,000 Android ready 2G/store-and-forward No patient PII

Synthetic district

Sehore Rural · 30 PHCs · 2 CHCs

Urgent centres4

Stock risk7d

Data confidence82%

Live recommendation: move ORS and malaria RDT kits from PHC Ashta Road to PHC Bhairavpur.

The challenge

Every required stream is handled without creating a new data-entry burden.

Stock

Voice note, register photo OCR and e-Aushadhi/DVDMS import become item-level stock events.

Footfall

OPD register photo, IVR count and HMIS reconciliation create daily aggregate demand signals.

Beds

CHC bed-board photo or IVR summary updates ward occupancy and referral pressure.

Attendance

Service continuity is inferred from prescriptions, OPD patterns and eSanjeevani sessions.

Tests

Lab shelf photo and voice note detect malaria RDT, Hb strip, pregnancy kit and reagent gaps.

Selected-round build

Phase 1 and Phase 2 are now split into product pages.

District dashboard

CMO-ready view of stock, beds, tests, footfall and service continuity.

AI intake workbench

Voice/OCR extraction, anomaly explanation and copilot answer in one flow.

Forecasting engine

Medicine, footfall, bed and test demand models with cold-start handling.

Transfer optimizer

Graph-based redistribution orders with expiry, buffer and transport constraints.

Working prototype

Voice note to district action

LLM confidence 91%

Demo language

FacilityPHC Bhairavpur

StockPending

FootfallPending

TestsPending

Lead-time aware

Stock-out queue

4 open

ORS · PHC Bhairavpur6 days cover · 21 day lead time

Critical

Hb strips · CHC Ichhawar11 days cover · ANC load rising

Watch

Amoxicillin · PHC Bilkisganj13 days cover · dengue season

Watch

Graph optimization

Redistribution order

min-cost flow

FromPHC Ashta Road ToPHC Bhairavpur

Move80 ORS packets VehicleBlock van

Sender remains at 24 days cover. Receiver stock-out avoided for 18 days. Batch expires in 78 days, so FEFO priority is satisfied.

Fair intervention

Centre health score

Stock resilience 42 Service continuity 78 Test readiness 39 Upstream dependency High

District copilot

CMO intervention brief

Top intervention: PHC Bhairavpur needs ORS and malaria RDT action before Monday. The stock issue is district-solvable through redistribution; the test issue requires district store confirmation.

Deployability and scale

Production path uses government systems plus Google Cloud services.

Capture

WhatsApp/IVR/PWA, register photo OCR, e-Aushadhi/DVDMS import, HMIS aggregate reconciliation.

AI layer

Speech-to-text, translation, OCR, LLM extraction, anomaly explanation and confidence scoring.

Forecasting

Gradient boosting for mature data, seasonal baselines for sparse PHCs, Croston/TSB for intermittent medicines.

Optimization

District graph with facility nodes, road-cost edges, FEFO lots, cold-chain constraints and min-cost flow.

Operations

Cloud Run APIs, Cloud SQL/PostGIS, Pub/Sub ingestion, Cloud Scheduler jobs, Looker/BigQuery analytics.

Privacy

Aggregate counts only, no ABHA or patient PII, raw images deleted after extraction, role-based district access.

What you hand in

Four submission items are embedded on this page.

Working prototype

This website runs the end-to-end voice-to-transfer-order demo flow.

Code repository

README, setup steps and CI/CD status are summarized below without leaving the page.

Pitch deck

The slide-by-slide pitch is included in the webform section.

Project write-up

Problem, architecture, tools, data, privacy and impact are included in the webform section.

Single-page hand-in

Complete submission webform

1. Project identity

Project name Track One-line pitch

ArogyaGrid turns PHC registers and WhatsApp updates into live forecasts, fair facility scores and actionable medicine redistribution orders.

Target users

PHC pharmacists, medical officers, CHC staff, district store managers, CMO/DPM offices and NHM district administrators.

2. End-to-end prototype flow

Capture

Hindi/Marathi voice note or register photo arrives from PHC staff.

Extract

LLM/OCR converts messy text into stock, footfall and test signals with confidence.

Forecast

Lead-time-aware demand model estimates days of cover and flags stock-out risk.

Optimize

District graph solver recommends a feasible transfer using buffer, expiry and transport constraints.

Approve

Draft transfer order goes to MO, district store and CMO for human approval.

3. Project write-up Problem

PHCs and CHCs face recurring stock-outs, unmanaged patient footfall, bed pressure, doctor service gaps and test downtime. Data exists across registers and siloed systems, but district administrators do not get early, actionable intervention signals.

Solution

ArogyaGrid is a multilingual AI operations layer over PHC/CHC workflows. It passively captures stock, footfall, beds, service continuity and test availability, then produces early warnings, forecasts, redistribution recommendations and under-resourced centre flags.

Zero-friction adoption

PHC staff do not fill a new MIS. They use WhatsApp voice notes, IVR calls, register photos and existing e-Aushadhi/DVDMS/HMIS/eSanjeevani data. The design works on low-cost Android phones and weak rural connectivity.

4. Technical architecture

Capture layer

WhatsApp/IVR/PWA, compressed register photos, e-Aushadhi/DVDMS stock imports, HMIS aggregate reconciliation and eSanjeevani session counts.

AI layer

Speech-to-text, translation, OCR, LLM extraction, confidence scoring, anomaly explanation and multilingual summaries.

Forecasting

Gradient boosting for mature demand history, seasonal baselines for sparse PHCs, Croston/TSB for intermittent medicines and cold-start matching by facility profile.

Optimization

Min-cost flow over PHC/CHC/warehouse graph with buffer stock, FEFO expiry, route cost, cold-chain and emergency reserve constraints.

Deployment

Hackathon: static website on GitHub Pages through GitHub Actions. Pilot: Cloud Run APIs, Cloud SQL/PostGIS, Pub/Sub ingestion, Cloud Storage lifecycle deletion and BigQuery/Looker monitoring.

Privacy

No patient PII or ABHA needed. Aggregate operational counts only. Raw register images are processed, redacted where needed and deleted after extraction.

5. Pitch deck content

01Title

Voice-first AI command centre for district PHC/CHC operations.

02Problem

Manual, delayed visibility causes stock-outs, downtime and referrals.

03Adoption

No new data entry; use voice, IVR, photos and existing systems.

04AI

Extract, forecast, detect anomalies and optimize redistribution.

05Demo

Voice note updates stock and creates transfer order.

06Pilot

30-50 facilities, CMO office, district store and NHM reporting.

6. Demo script and hand-in status Three-minute demo

0:00 Show 32-facility district dashboard. 0:30 Process Hindi voice note from PHC Bhairavpur. 0:50 Extract ORS, paracetamol, malaria RDT and OPD fever count. 1:10 Trigger stock-out warning with 21-day lead time. 1:25 Generate redistribution from nearby PHC. 1:50 Mark transfer order ready for CMO review. 2:10 Ask district copilot for Monday interventions.

Working prototype included Code and setup included Pitch deck content embedded Project write-up embedded CI/CD deployment configured No outbound document redirects

Ready for review.