Government of Maharashtra | Project PRAGATI
Deep-dive System Architecture
Five interoperable layers from identity ingestion and clinical systems through AI decision support to state analytics, security and compliance.
Draft for discussion | October 13, 2026
Architecture at a Glance
Five layers, shown as a vertical stack. Each layer is described in its own section below. AI outputs are decision support only and always require clinician sign-off.
L5 - Security, Audit and Compliance
ABDM compliance | ISO/IEC 27001 alignment | DPDP Act 2023 | audit trail | role-based access | encryption
L4 - State Health Data Exchange and Analytics
FHIR repository | SNOMED CT / LOINC terminology | population surveillance | bed / ICU / drug registry
L3 - AI Clinical Decision Support and Triage
radiology AI triage | predictive triage | maternal-child scoring | LLM/NLP summariser (human in the loop)
L2 - Clinical Systems Core
HMIS (OpenHMIS ref) | LIS | Cloud PACS + DICOM router | telemedicine (eSanjeevani)
L1 - Ingestion and ABHA Identity
ABHA creation | linking | PHR app | e-KYC flow
Layer 1 - Ingestion and ABHA Identity
Identity foundation
The front door for patients and frontline workers: every person gets an ABHA-backed identity and their records are linked to it through the ABDM ecosystem, with an e-KYC flow to establish identity.
L1 / Ingestion and ABHA Identity
Identity and enrolment
- 1
ABHA creation via e-KYC at a facility or self-service kiosk
- 2
Health records linked to the ABHA through the ABDM gateway
- 3
Patients access records and grant consent through a PHR app
Note:ABHA flows reference the Ayushman Bharat Digital Mission. Confirm the exact gateway flows against sandbox.abdm.gov.in.
Layer 2 - Clinical Systems Core
The transactional spine
The clinical systems that admit, treat and investigate patients. OpenHMIS is noted as an open-source HMIS reference option; the LIS, Cloud PACS and telemedicine components integrate with it.
HMIS
Hospital management + EHR. OpenHMIS as an open-source reference option.
LIS
Laboratory information system with result workflows and LOINC coding.
Cloud PACS
DICOM archiving with a DICOM router to route studies to reads.
Telemedicine
Integration with eSanjeevani for virtual consults.
Layer 3 - AI Clinical Decision Support and Triage Engine
Decision support, never autonomous
AI augments, clinicians decide. Every AI output in this layer is triage or decision support only and requires clinician validation before any action. No accuracy figures are stated because they are not sourced.
Radiology AI triage
Chest X-ray and CT triage supporting prioritisation; outputs are model flags for clinician review.
Predictive triage
Risk-based prioritisation at triage desks using structured intake data.
Maternal and child scoring
High-risk scoring for maternal and child health based on antenatal and growth data.
LLM / NLP summariser
Clinical text summarisation with mandatory human-in-the-loop clinician validation.
AI outputs are decision support only and require clinician sign-off. Automated action on AI output alone is not permitted.
Layer 4 - State Health Data Exchange and Analytics
Interoperability and population insight
A standards-based repository and terminology layer that underpins population health surveillance and the live hospital bed, ICU and drug registry.
| Component | Function |
|---|---|
| FHIR repository | Longitudinal records in FHIR R4 |
| SNOMED CT / LOINC terminology services | Standardised codes and value sets |
| Population health surveillance | NCD and cancer screening analytics, call-recall |
| Hospital bed / ICU / drug registry | Real-time capacity across the five hospitals |
Layer 5 - Security, Audit and Compliance
Trust by design
Security controls that satisfy ABDM compliance, align to ISO/IEC 27001 and meet the DPDP Act 2023, with an immutable audit trail, role-based access and encryption at rest and in transit.
- ABDM compliance for identity, consent and data exchange
- ISO/IEC 27001 alignment for the information security management system
- DPDP Act 2023 consent and data protection obligations
- Audit trail design capturing who accessed what, when and why
- Role-based access control (RBAC) across all user groups
- Encryption of data in transit and at rest
Standards Map
FHIR .......... HL7 FHIR R4 (profiles per ABDM IG) DICOM ......... radiology imaging and PACS exchange LOINC ......... laboratory observations and diagnostic codes SNOMED CT ..... clinical findings, conditions and procedures ICD-11 ........ morbidity and mortality coding HL7 v3 ........ legacy administrative / gender codes ABDM .......... identity, consent and data exchange via gateway
Note:Standards align to the FHIR Implementation Guide for ABDM (nrces.in/ndhm/fhir/r4). Scope and architecture are indicative proposals for discussion.