Community Platform v2.0
A full-platform rebuild that cleared 3 years of accumulated technical debt, reduced page load times by 71%, and grew monthly active users from 38K to 120K delivered in 20 weeks without a single hour of unplanned downtime.
The Challenge
A platform that had outgrown the codebase it was built on.
A UK-based creator community platform had grown to 38,000 monthly active users on a monolithic PHP/jQuery codebase originally built in 2021 by a team that had since departed. Over three years, the platform had accumulated enough technical debt to make every new feature a multi-week negotiation with legacy code.
Average page load time had crept to 7.2 seconds. The database had no query optimization strategy every user profile load triggered 40+ unindexed queries. The deployment pipeline was a manual FTP upload. And the real-time messaging feature, which users depended on daily, was built on polling rather than WebSockets, creating significant server load at peak hours.
The client's churn rate was climbing. User feedback consistently pointed to performance. The product team had a 12-month feature roadmap they couldn't execute on the existing stack. They needed a complete rebuild but couldn't afford to go dark for months while we built it.
Project Parameters
Industry: Creator Economy / Online Community
Client: Confidential UK-based creator platform
Timeline: 20 weeks, discovery to full production cutover
Our Approach
Rebuild in parallel. Cut over without downtime. Ship the roadmap immediately.
We ran v1 and v2 in parallel throughout the entire build, using a strangler-fig migration strategy. The new Next.js frontend consumed the same data layer as v1 during development meaning we could run real user traffic on v2 behind a feature flag weeks before the final cutover, surfacing real-world edge cases without exposing users to risk.
Database architecture was rebuilt from the ground up moving from a single overloaded PostgreSQL instance to a properly indexed, read-replica-backed setup with Redis caching for hot data. The 40+ queries per profile load became 3 with proper relational design and eager loading.
Real-time messaging was re-engineered with WebSockets via Socket.io, eliminating the polling load that had been hammering the server during peak hours. The deployment pipeline moved to GitHub Actions with Kubernetes on AWS EKS giving the client's team one-command deployments where they previously had none.
The Solution
Three architectural decisions that changed everything.
Next.js App Router + Edge Rendering
The new frontend was built on Next.js 14 with the App Router, using React Server Components for content-heavy pages and client components only where interactivity required it. Community feed pages render at the edge via Vercel's CDN meaning users in London, Lagos, and Los Angeles all see sub-2-second first loads regardless of origin server location.
Database Redesign with Read Replicas
The PostgreSQL schema was redesigned with proper normalisation, composite indexes on all high-frequency query paths, and a dedicated read replica for analytics and feed generation. Redis caches user session data, follower counts, and hot content reducing database pressure by 80% during peak traffic. The result: a profile that previously required 40+ queries now resolves in 3.
Strangler-Fig Migration Strategy
Rather than a big-bang cutover, we ran v2 in shadow mode routing a percentage of real traffic to v2 behind feature flags, logging discrepancies, and resolving edge cases for 6 weeks before the final switch. The cutover itself took 4 minutes. No user-facing downtime. No emergency rollbacks. Just a clean transition that the engineering team has never had to revisit.
The Results
A platform built to scale numbers that proved it immediately.
71%
Faster Page Loads
7.2s → 2.1s average
3×
Monthly Active Users
38K → 120K in 6 months
80%
Database Load Reduced
At peak traffic hours
0 min
Unplanned Downtime
Entire 20-week build + cutover
Tech Stack
Next.js 14 at the edge, PostgreSQL with read replicas, Redis for hot data, and Kubernetes on AWS EKS a stack built for the next five years, not just the next sprint.
Frontend
Next.js 14 (App Router) · React Server Components · TypeScript · Tailwind CSS · Socket.io client · Zustand · SWR
Backend & Data
Node.js · Express · PostgreSQL (primary + read replica) · Redis · Socket.io · Prisma ORM · S3 for media storage · Cloudfront CDN
Infrastructure & DevOps
AWS EKS (Kubernetes) · Terraform · GitHub Actions CI/CD · Vercel Edge Network · Datadog monitoring · PagerDuty alerting
Carrying technical debt that's slowing you down?