Edward Shen

Lifecycle campaigns repeatedly combined latest user traits, historical events, and expiring assets. Operations initially assembled those audiences manually; the platform already had transactional services and a TiDB/TiCDC/Kafka/Flink data path, but marketing needed reusable audience definitions and automated email tasks rather than campaign-specific filters.

I wrote the design document and early core demo for predicate expansion, SQL compilation, segment refresh, and the reach handoff. A colleague implemented the production services with my guidance.

The Go backend platform had been established earlier in 2025 and eventually carried eighteen Go domain services. Product teams could ship features, but operational questions still required ad-hoc production queries. Dashboards were slow, and linked views could not be trusted to refresh from a consistent snapshot.

Four Go engineers had to build the data platform alongside feature delivery. We could not operate a separate OLAP database, sync pipeline, and dedicated data engineering function, so we built TiCDC → Kafka → Apache Flink → analytical tables in TiDB, served by QueryHub, a self-developed SQL-template service.

Four Go engineers built eighteen domain services for a consumer transaction platform with weekly live-ops releases — time-boxed in-game events and promotions whose cadence set the delivery rhythm. To keep integration work manageable, we used a single API DSL to generate service conventions, Protocol Buffers contracts, and Swagger/OpenAPI documentation.