Case studies

How the work actually went

Cloud Architect · Technical Lead · GCP Professional Cloud Architect · GCP + AWS

Six projects from CNET's platform teams and one AI platform lab: the problem, the architecture, and how we verified the result.

• Fixing a cross-cloud slowdown with batched lookups

  After a datastore move from GCP to AWS slowed pricing feeds 4–9×, I traced the bottleneck to cross-cloud network round trips rather than the database and batched the lookups—cutting feed runs ~70% below pre-migration times.

• Re-platforming product search from Solr 7 to Solr 9

  I designed and led the consolidation of nine legacy Solr 7 cores into three Solr 9 cores on AWS EKS (~30M documents, no planned downtime), building a 15-point verification suite to catch silent import and replication failures.

• Moving CNET's CMS and content API from Docker Swarm to GKE

  I led CNET's CMS and content API migration from Docker Swarm to GKE—owning the cost model, replica sizing, in-cluster Varnish cache tier, and phased evening cutovers with zero rollbacks.

• Carving TV Guide and Metacritic out of a shared platform

  As the sole engineering contact for TV Guide and Metacritic, I led their technical separation from a five-brand shared CMS into standalone, buyer-ready platforms.

• Self-service integration environments for outside teams

  I built a self-deploying Cloud Run integration branch for our WordPress VIP migration vendor—providing a live, login-gated endpoint that scaled to zero while main kept shipping—and reused the pattern to onboard RetailMeNot end to end.

• Making API tests the AI agent's verify step

  I built the team's Bruno API test repository with guardrails for AI coding agents—read-only by default, runtime secret injection, and a single pass/fail verdict per environment—making live API verification the final gate before a pull request.

• Testing an LLM gateway's controls instead of trusting its config

  In a LiteLLM lab, I tested provider fallbacks, Redis caching, budgets, MCP access, and spend tags—uncovering a custom-auth bypass that disabled team-level controls and a route that silently dropped cost attribution.