The problem
Lumen's clinician platform served 4,000+ daily active users on a four-year-old monolith. Page loads were exceeding two seconds, the team was spending most sprints on incident response, and a planned third-party integration was blocked by the existing schema.
What we did
We ran a two-week architecture audit, then proposed an incremental rebuild — strangler-fig pattern, no big-bang migration. We moved the read paths first behind a new typed API layer, introduced read replicas and caching, and progressively cut over write paths module by module.
The result
P95 latency dropped from 1.2 seconds to 180ms. The team shipped three previously-blocked modules during the migration itself. There have been zero production incidents in the six months since launch.