Rebuilding a global photography competition platform for 25 million users.

The Challenge

The existing platform had reached its architectural ceiling. As the user base grew beyond 20 million, the team encountered cascading performance issues: slow page loads during peak contest periods, image processing bottlenecks, and a monolithic codebase that made feature development increasingly risky and slow.

Contest launches generate extreme, concentrated load — thousands of photographers uploading high-resolution images while millions browse and vote simultaneously. The architecture had to scale horizontally on demand and degrade gracefully under pressure, with zero tolerance for downtime during peak windows.

The platform also required multi-tenant isolation at scale, an optimized image processing pipeline handling millions of photographs, layered voting integrity measures, and a mixed rendering strategy delivering both performance and SEO across billions of pageviews.

Our Approach

We designed a serverless-first architecture on AWS, replacing the monolith with a modular backend deployed via infrastructure-as-code. The frontend was rebuilt with a mixed rendering strategy — SSR, ISR, and SSG — optimized per route type for the ideal balance of freshness and performance.

The most distinctive decision was a multi-layered caching strategy, with each layer optimized for a different access pattern — from edge CDN for static assets, through rendering and database acceleration layers, down to application-level and in-memory caches. This architecture absorbs the vast majority of traffic before it reaches application code.

The image pipeline processes uploads through server-side transformation — resizing, format optimization, and metadata extraction — before delivering optimized variants via CDN. Real-time features use multiple communication channels to ensure participants see voting results and contest updates instantly regardless of their network environment.

The Results

A platform serving 25 million active users, rebuilt from scratch and maintained by a 3-engineer team — enabled by aggressive automation, serverless infrastructure, and a caching architecture that absorbs the majority of traffic before it reaches application code.

Sub-second response times during peak traffic. Zero downtime during contest launches. A modular codebase where features can be developed and deployed independently, dramatically reducing the risk and cycle time of every release.