Client — fintech broker (CFD/FX), 12K daily active traders. Legacy dashboard polled REST every 3s — stale prices, client complaints, risk of orders on outdated quotes. Peak load: 8K concurrent WebSocket connections during market open. Scope: real-time web dashboard — live quotes, order book depth, portfolio P&L, sub-500ms UI updates. Failover if feed disconnects; mobile-responsive. Implementation: • WebSocket gateway (horizontal scale): sticky sessions, Redis Pub/Sub fan-out from price feed; • backend: NestJS + Redis Streams buffer, normalizer for 40+ instrument feeds; • frontend: React + lightweight charts, virtualized order book, reconnect with snapshot sync; • backpressure: drop non-critical ticks under load, always deliver bid/ask; • auth: JWT short-lived + refresh; session revoke on password change; • monitoring: connection count, lag p99, alert if lag > 1s; • load test: 10K connections, 50K msgs/sec burst; auto-scale K8s HPA; • compliance: audit log of viewed instruments (regulatory request). Stack: React, TypeScript, WebSocket, Redis, Node.js, PostgreSQL, Kubernetes, Grafana. Acceptance: p99 tick-to-UI < 400ms under load test. Result: support tickets "wrong price" −85%; session duration +30%. NDA: broker not disclosed.