System Architecture

System Architecture Overview

Lume Destinations runs on a serverless-first architecture deployed on Google Cloud Platform (GCP). The platform is built with Next.js 15, React 19, and PostgreSQL, optimized for high concurrency and low latency.

Tech Stack

Infrastructure

Compute (Google Cloud Run)

• Dev Environment: lume-destinations-dev (deploys on all pushes)
• Production: lume-destinations-prod (deploys on git tags v*.*.*)
• Region: us-central1
• Auto-scaling: 0-100 instances based on CPU

Database (Cloud SQL PostgreSQL)

• Instance: gen-lang-client-0870850972:us-central1:lume-db
• Connection: Private IP (via Cloud SQL Proxy in production)
• ORM: Prisma with multi-target binaries

Storage (Google Cloud Storage)

• Bucket: assets.infillcapitalpartners.com
• Content: Images, videos, documents
• Delivery: Direct HTTPS (no CDN currently)

Deployment Pipeline

Continuous deployment via Google Cloud Build (cloudbuild.yaml):

1. Build: Multi-stage Docker build (node:20-alpine)
2. Push: Container to GCR with tags $SHORT_SHA + latest
3. Deploy Dev: Always deploys to development
4. Deploy Prod: Conditional on git tags matching v[0-9]+.[0-9]+.[0-9]+

Build time: ~3-5 minutes

Route Map

Public Pages (44 total)

Homepage, locations, pod types, experiences, careers, blog, press, contact, FAQs, policies, etc.

Protected Pages (Requires Auth)

• User Dashboard: Profile, trips, rewards, referrals, wallet, concierge
• Admin Portal: Analytics, loyalty management, email templates

Data Models

The database schema (via Prisma) includes:

• Users: Auth, membership tiers (Bronze → Founder), points, referral codes
• Bookings: Status workflow, source attribution, points earning
• Loyalty: Reward transactions, catalog, tier history, Apple Wallet passes
• Referrals: Coupons, activity tracking, fraud detection
• Payouts: Methods, requests, approval queue

Performance Bottlenecks

1. LCP: ~2.5s due to large hero images without CDN edge caching
2. Image Weight: On-demand Next.js optimization causes cold start penalty
3. Database: No connection pooling configured (+50-100ms overhead)
4. Cold Starts: Dev environment scales to zero (~1-2s first request)

Recommended Quick Wins