Architecture Briefing

Prudential Health OS is an event-driven insurance platform.

Members, agents, hospitals, claims ops, payments, and leadership intelligence stay in sync through a governed claim-event backbone.

Event-drivenAI governedNHCX-readyAudit-firstReal-time ops
System at a glance
Experience surfaces subscribe to the same claim truth.
Member
Agent
Hospital
Ops
Leadership
API + Identity
Claim Event Bus
Domain Services
AI + Rules
Data + Rails
Claim state
Replayable
AI role
Recommends
System role
Decides
Product loops

The architecture is organized around four business loops

The same platform services power member engagement, cashless automation, agent productivity, and executive visibility.

System overview

A layered platform with one event backbone

The diagram shows ownership boundaries rather than implementation trivia. Each layer has clear contracts and emits observable state changes.

Experience
Member, agent, hospital, ops, leadership
Next.js surfacesWhatsApp/RCSEmbedded SDK
Edge
API gateway, identity, partner trust
OIDC/OAuthmTLSWAF/rate limits
Backbone
Kafka claim saga and outbox events
claim.* topicsCDCidempotency
Domain
Bounded-context services
PolicyClaimsProviderBilling
Intelligence
Rules, document AI, clinical NLP, LLM assist
Policy DSLCitationsFraud score
Rails
Payments, regulators, hospitals, analytics
NHCXNEFT/RTGSIRDAIWarehouse
Cashless claim saga

The core journey is an auditable event stream

Every service sees the same claim ID, emits its own event, and can replay the saga for audit, recovery, analytics, and member transparency.

Step 1
Hospital pre-auth
FHIR packet
claim.preauth.received
Step 2
Eligibility
Policy + member
claim.eligibility.verified
Step 3
Document AI
OCR + classifier
claim.docs.extracted
Step 4
Rules engine
Deterministic DSL
claim.rules.evaluated
Step 5
AI assist
Cited recommendation
claim.ai.recommended
Step 6
Adjudication
Auto or L2 review
claim.adjudicated
Step 7
Settlement
Bill reconciliation
claim.settled
AI governance

AI is powerful, but not sovereign

The production pattern is deterministic boundaries plus AI assistance. That gives speed without losing explainability or regulatory control.

Rules first

Hard eligibility, waiting periods, sub-limits, and exclusions run before any model.

AI recommends

Document AI and LLMs extract, summarize, cite, and recommend. They do not directly approve.

Human-in-loop

Low-confidence, high-value, or exception cases move to L2/L3 review with provenance.

Audit + model ops

Every field stores source, confidence, version, reviewer, and decision trace.

Decision contract
AI recommends. Governed services decide.
Input
Hospital packet, policy, clinical notes, tariffs
Output
Recommendation, rationale, citations, confidence
Guardrail
Rules engine, thresholds, reviewer queue
Audit
Model version, source document, field provenance
Integration map

The platform sits between care providers, regulators, comms, and money rails

Boundaries are explicit: partner adapters handle external variability while the internal event model stays stable.

NHCX / ABDM
Hospital EMR / TPA
ABHA / DigiLocker
NEFT / RTGS / IMPS
Claim Event Backbone

Canonical events, adapters at the edge, audit trail in the core.

WhatsApp / RCS / SMS
IRDAI reporting
CRM / agent systems
Re-insurer bordereau
Reliability, security, compliance

The non-negotiables are built into the platform model

For health insurance, speed only matters if the result is traceable, secure, and explainable.

Identity and trust

OIDC, mTLS, partner certificates, device binding.

Data protection

AES-256 at rest, TLS 1.3 in flight, KMS-backed keys.

Data residency

Primary India region, DR in-region, PHI boundary controls.

Observability

OpenTelemetry traces, business SLIs, claim ID propagation.

Recovery

Replayable events, idempotent payouts, DLQs, backfills.

Auditability

Every decision links to source, actor, rule, model version.

Prudential Health
Why this architecture wins

It combines frequent member engagement, cashless automation, governed AI, lower ops load, and regulator-ready auditability without coupling every product surface to every service.

Open leadership view