At EPAM, I improved Kafka message processing under load by routing recoverable failures to a dead-letter path, retrying transient errors, and isolating persistent failures so healthy traffic kept moving.
I also accelerated the slowest API responses from roughly 600 ms to 200 ms at p95 by profiling read paths, adding an index, and fixing cache invalidation. For a shared reference-data service, I cached hot reads and reached about a 90% cache hit rate on hot paths.
Before EPAM, at ELMA365, I built a shared object model and domain APIs for ELMA365 Projects (Gantt), helping low-code integrators adopt and extend the solution with less custom work. I also implemented a shared event handler to record platform changes and publish them on one path.
