— Selected work
Recycling at the source
Migrated a recycling marketplace from shared hosting to a scalable cloud infrastructure, resolving chronic performance issues and enabling the product team to ship again.
A recycling materials marketplace was operating on shared hosting infrastructure it had outgrown two years earlier. Peak-hour response times were routinely above eight seconds. The development team had stopped deploying because every deployment risked taking the site down. The product roadmap was frozen.
The problem
Shared hosting creates a specific category of production problem: you don’t control the environment, you can’t instrument it properly, and you can’t reproduce the failures locally because your local setup doesn’t match. Debugging is guesswork.
The immediate symptom was slow response times. The root cause was a combination of under-resourced MySQL, no opcode caching, no Redis for session or cache storage, and a deployment process that involved FTP and manual cache clearing. Every layer of the stack had a problem.
What we built
The migration was done in two phases. Phase one was a like-for-like lift: same application code, new infrastructure. AWS EC2 for the application layer, RDS for MySQL with properly sized instance and tuned configuration, ElastiCache for Redis sessions and Magento cache, Cloudflare in front for DDoS protection and static asset caching.
Phase two was the stability work. A Docker-based local development environment that matched production meant developers could reproduce issues locally for the first time. A CI/CD pipeline replaced the FTP deployment process. Automated smoke tests verified the site was functioning after each deploy.
Redis-backed full-page cache configuration was tuned for the marketplace’s traffic patterns — product pages cached for 4 hours, category pages for 1 hour, customer-facing pages excluded. Cloudflare cache rules were layered on top for static assets and product images.
Outcomes
Peak-hour response times dropped from eight seconds to under 600ms within two weeks of the infrastructure migration. The development team shipped their first feature in three months within a week of the CI/CD pipeline going live.
The infrastructure costs were higher than shared hosting — about 3x monthly spend — but the team was able to quantify the lost GMV during slow periods and make the case internally. Within six months of the migration, conversion rate on product pages had recovered to a level that more than covered the infrastructure cost increase.