The platform cost $44,000 a month to run.
Now it costs $867.

The platform ran on a single outside vendor — development and hosting both — that also owned the source code, priced at $1M to buy. I recommended rebuilding in-house instead, and led the exit against two contractual deadlines.

I reverse-engineered the vendor's undocumented stack (VMware, Docker Swarm, Traefik, NFS, MinIO, Redis, Postgres) to plan the migration, provisioned a transitional AWS environment as Terraform modules to keep the legacy application serving through the rebuild, then decommissioned all 75 resources afterward — tearing it down only after restoring its offsite backup into a clean Postgres and diffing 3.23M rows against production.

A complete rebuild on Next.js and Payload CMS with Postgres: native authoring with scheduled publishing and draft preview, adaptive-bitrate video on an in-house encoder, self-hosted transcription, full-site search that reindexes on publish, reader accounts and comments, newsletter automation, and first-party analytics.

I sized the replacement from measurement rather than assumption — mining the legacy load balancer's logs for the true worst traffic minute and the database for the real signed-in share — then load-tested to 2.4× that peak with zero errors before cutting over. The live cutover between two entirely different stacks cost two roughly thirty-second interruptions, on a scripted DNS and cache-rule sequence with a rollback I had dry-run first. I needed it: the first flip exposed a proxy bound only to the preview hostname, so I rolled back and re-flipped clean.

Event-driven ingestion across WinRed, Revv, and RaiseDonors built on AWS Lambda behind SQS queues, with dead-letter queues for failed deliveries and a DynamoDB event log so any run can be replayed. Idempotent processing throughout, plus the HubSpot data model and automation behind recurring gifts, cancellations, and the full supporter lifecycle.

I later recovered the RaiseDonors-to-HubSpot pipeline through a breaking third-party API change, keeping supporter records and recurring revenue intact across the vendor transition.

Led the rebuild from legacy WordPress to a Next.js and Strapi headless architecture, paired with a junior developer on a four-person engineering team while two engineers held the legacy site in production.

I built custom Dockerized ETL pipelines that migrated a twenty-year archive — 100K+ articles plus media assets and taxonomy — into the new stack, with an incremental sync so ongoing update cycles pulled only changed records instead of re-running a full import. I also implemented role-based access control alongside donation, petition, and crowdfunding flows the legacy site could not support.