A legacy Python platform owned identity, payments, integrations, and administration. I built new Go services while maintaining old production behavior and standardized how mixed-language services ship.

Service extraction could not become organization-wide migration.

Mobile, web, backend, and integrations already depended on legacy token shapes, configuration names, and response envelopes. Breaking those contracts would move risk from one service into every caller.

Make new systems absorb old contracts.

1. Compatibility firstNew services accepted legacy authentication objects, normalized configuration aliases, and unwrapped old proxy envelopes.
2. Retries safe by designIdempotent registration and account-creation boundaries prevented duplicate work.
3. Independent deliveryEach service received tests, build pipeline, deployment lane, and smoke probe.
4. One fleet contractRust, Go, and Python services used digest-pinned images, workload-identity secrets, and common post-deploy checks.

Worked both sides of extraction.

I authored two Go services from scratch and major parts of three more while continuing to maintain Python monolith paths they replaced. This let compatibility decisions follow real caller behavior, not an idealized rewrite.

Rust services were written by another engineer. My contribution there covered deployment, configuration, and CI, not service authorship.

Independent build and test cycles, without unsupported performance claims.

Six Go services carried 578 test functions across 76 files. Around fifteen services shared one delivery pattern. Extracted services had staging evidence; monolith remained production system of record. No before-and-after latency or infrastructure-cost benchmark existed, so none is claimed.