convez.ai
AgenteDevOps & InfraestruturaAvançado
dkds

Claims Intelligence Platform

dkds/claims-intelligence-platform

Microserviços event-driven para processamento de sinistros veterinários com CQRS

Ver no GitHub
InicialPluginScore · de 0 a 100

Plataforma event-driven de microserviços para processamento de sinistros de seguro veterinário, implementada em Java, Go, Node.js e React com Kafka. Demonstra arquitetura limpa, domain-driven design e CQRS, com cada serviço possuindo banco de dados isolado e lógica de negócio em código.

Instalação

$npx convez-ai dkds/claims-intelligence-platform

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 3 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Construir arquitetura de microserviços event-driven em produção
  • Implementar CQRS e eventual consistency em sistemas distribuídos
  • Desenhar pipelines de processamento complexos com domain-driven design

Veredito

Projeto de referência sólido para quem quer estudar ou implementar microserviços com separação clara de responsabilidades. Código bem estruturado e documentação clara sobre topologia de eventos, mas requer experiência com arquitetura distribuída para aproveitar.

README

A tradução para o português deste README ainda está na fila.

Claims Intelligence Platform

A portfolio project demonstrating a clean, event-driven, polyglot microservices architecture for processing veterinary insurance claims. The platform deliberately corrects two common anti-patterns: a database shared across services, and business logic encoded in the database.

Domain and all data are generic and synthetic.


Overview

The platform is a third-party administrator sitting between veterinary clinics and an insurer. Claims arrive through two intake paths and converge into one pipeline:

Vet logs session ──► Clinic manager verifies ──► Session-derived claim ──┐
                                                                          ├──► Fraud score ──► Triage ──► Adjudication ──► Insurer
Clinic manager submits manually ─────────────────────────────────────────┘
  • Session-derived: a vet logs a treatment session → clinic manager verifies → auto-generates claim lines.
  • Manual: clinic manager submits a claim directly (meds, transport, boarding), bypassing sessions.
  • Convergence: every claim is fraud/risk-scored, then triaged — verified low-risk claims auto-adjudicate; manual or high-risk go to adjuster review — then submitted to the insurer. Verified sessions also trigger a vet payment.

Architecture Principles

PrincipleHow it is applied
Database per serviceEach service owns its store privately; no service reads or writes another's database.
Logic in codeBusiness rules live in a rich domain model and rules engine — never in stored procedures or schema.
Event-drivenServices communicate via domain events on Kafka; the transactional outbox ensures reliable publishing.
CQRSA single Projection service is the only writer of the read-only view store; everything else reads through the BFF.
Eventual consistencyDecisions read strongly-consistent local data; display and reporting tolerate eventual consistency.

Services

ServiceStackDatabaseRole
Enrollment & PolicyJava / Spring BootPostgreSQLMaster data: clinics, pets, vets, catalogue, policies
SessionsJava / Spring BootPostgreSQLTreatment session logging and verification
ClaimsJava / Spring BootPostgreSQLClaim assembly, validation, triage, adjudication, state machine
Fraud DetectionGo—Rules-based fraud/risk scoring
ProjectionNode.js / NestJSMongoDBBuilds and owns the CQRS read model
BFF / GatewayNode.js / NestJS—Auth, aggregation; reads view store, dispatches commands
Clinic WorkbenchReact + Vite + TypeScript—Verify sessions, submit claims, adjuster queue, dashboard

Event Topology

Services communicate through domain events on Kafka. All events use a standard envelope carrying an event ID (for idempotency), correlation/causation IDs (for tracing), and a typed payload.

TopicProducerKey events
cip.enrollment.v1Enrollment & Policyclinic.registered, pet.enrolled, vet.approved
cip.sessions.v1Sessionssession.logged, session.verified
cip.claims.v1Claimsclaim.assembled, claim.adjudicated, claim.routed-to-review
cip.fraud.v1Fraud Detectionclaim.fraud-scored
cip.payments.v1Payment (Tier 2)payment.completed
cip.submissions.v1Insurer Submission (Tier 2)claim.submitted, claim.accepted

Local Development

The dev environment runs entirely inside a dev container. All infrastructure is managed by Docker Compose.

Infrastructure hostnames:

ServiceHost
PostgreSQLpostgres:5432
Kafka (Redpanda)redpanda:9092
Redpanda Consolelocalhost:8085
MongoDBmongo:27017
Redisredis:6379

Run a service:

# Java services
cd services/enrollment && ./mvnw spring-boot:run
cd services/sessions   && ./mvnw spring-boot:run
cd services/claims     && ./mvnw spring-boot:run

# Go
cd services/fraud-detection && go run ./...

# Node services
cd services/projection && pnpm run start:dev
cd services/bff        && pnpm run start:dev

# Frontend
cd web/clinic-workbench && pnpm run dev

Run tests:

./mvnw test          # Java
go test ./...        # Go
pnpm test            # Node / frontend

Build Phases

Tier 1 is built incrementally — something runnable at every step:

PhaseFocusStatus
0Foundations: monorepo, dev container, infra, service stubsComplete
1Enrollment & Policy: REST API, master data, vet approvalComplete
2Sessions + Kafka: first event flow, session verificationComplete
3Claims: assembly, rules engine, state machineComplete
4Fraud Detection (Go) + triage routingComplete
5Projection + BFF: CQRS read modelUpcoming
6Clinic Workbench: end-to-end UI—
7Hardening: outbox, schema registry, idempotency, auth—

Documentation

Ver o original no GitHub

Bom para

Relacionados