Projects

Case studies

More projects

Problem: A multi-tenant host serving about a dozen client sites on legacy LAMP stacks, with part of the production estate on GCP.
What I did: Moved about 12 legacy sites from LAMP 5 to LAMP 8. Migrated the production estate from GCP to DigitalOcean. Introduced blue/green deployments. Gave each deploy user a chroot SFTP jail with bind-mounted docroots, so a user sees only their own sites. Moved sites that depend on .htaccess rewrites from Nginx to Apache. Manages Cloudflare Origin certificates by hand so certbot never overwrites them.
Result: About a dozen sites on current PHP, production off GCP, deployments without downtime.

Problem: A site that could only be updated over SFTP, by hand.
What I did: GitLab CI pipelines that deploy into a chroot SFTP jail, with bind mounts and a permission sync so the deploy user and the web server always agree on ownership.
Result: Deploys happen from a git push instead of a manual upload.

Problem: Application-consistent snapshots on GCP were failing silently under continuous disk I/O.
What I did: Found the silent failures, switched to crash-consistent snapshots encrypted with customer-managed keys (CMEK), on a schedule.
Result: Backups that run on schedule instead of failing without a trace.

Problem: A backend service built on the Claude API was spending far more on tokens than it needed.
What I did: Traced the cost to a large conversation-history window and bloated tool results, then shaped tool outputs to the fields actually used, shortened the history window and added prompt caching.
Result: Token spend brought under control without changing what the service does.

Problem: Company email needed moving to Microsoft 365 without losing deliverability.
What I did: Set up Microsoft 365 Business Basic and completed the full DNS cutover — MX, SPF, DKIM and DMARC — plus mailbox warm-up guidance.
Result: MX, SPF, DKIM and DMARC all passing.

Problem: Two seasonal businesses depended on a booking platform on legacy VPS hosting and paid SaaS tools, and it had to survive peak holiday surges.
What I did: Moved it to DigitalOcean droplets and containerised the custom LAMP booking engine with Docker Compose, separating database, application and web tiers. Monitoring with sysstat and custom uptime webhook alerts.
Result: Ran in-house through peak seasons.

What I did: Moved clients from shared hosting to managed VPS: an Nginx caching reverse proxy in front of Apache/mod_php, clients separated by chroot jails and their own UNIX users, cron-wrapped certbot renewing SSL across dozens of virtual hosts, and automated MySQL dumps with retention.

What I did: Ran bare-metal CentOS 6 servers for gaming communities that needed uptime and low latency. Bash scripts bootstrapped environments and shipped daily tarball backups to an off-site FTP server. iptables rate-limiting and connection tracking held off UDP/TCP floods, and sysctl tuning kept the network stack fast.