How priyankabedekar-ms/fintech-case-console works
Project type and purpose
• This is a full-stack fintech case-resolution console (“Zeta Sentinel”) comprising:
• A React/TypeScript single-page UI (web/)
• An Express/TypeScript API (api/)
• A PostgreSQL relational store and Redis for rate-limiting & job queuing
• Docker Compose for local development and orchestration
• Its purpose is to ingest alerts, run multi-agent AI triage, stream recommendations via SSE to the UI, execute user-driven actions (e.g., freeze card), record audit trails, and expose Prometheus metrics.
File structure
• Top-level:
• api/ – source, Prisma schema, migrations, Dockerfile
• web/ – React app, Vite config, Dockerfile
• docker-compose.yml – orchestrates api, web, postgres, redis, prometheus
• docs/ – ADRs, fixtures, Postman collections
• api/src/:
• routes/ – HTTP endpoints (triage, actions, alerts, insights)
• agents/ – orchestrator drives sub-agents (profile, riskSignals, decision)
• middleware/ – rate limit, security, metrics
• lib/ – Redis client, PII redactor, metrics helpers
• eval/ – offline evaluation suite
• web/src/:
• components/, pages/, hooks/ – UI layers, SSE client, API wrapper
• styles/, assets, entry & HTML
README insights
• Request flow: POST /triage → orchestrator → SSE → UI; POST /action/* → audit event
• Key architectural decisions: server-sent events, token-bucket rate limit, PII redaction at edge, event sourcing for audit, idempotency, Prisma ORM