Danila Koshevoi

Building a giga-mega Trading API

Payment Services Team • Extracted payment logic from a monolithic system and built a scalable payment microservice from scratch using Golang, CockroachDB, Clickhouse and Kafka, now handling 60k+ transactions/day with 99.9% uptime. • Developed a robust subscription lifecycle manager handling trials, renewals, grace periods, proration logic, and automatic retries, improving subscription retention by 12% and reducing support tickets related to billing issues by 40%. • Enhanced the mailing microservice, improving send throughput from 100k to 150k emails/hour by optimizing processing with Kafka and Redis. • Added end-to-end tests and service-level dashboards (Prometheus/Grafana) with SLO-based alerts, reducing incident investigation time.

Buyer Experience Team • Launched and managed 23 A/B tests, with 12 tests successfully concluding and being fully deployed, resulting in optimized features and improved user experience. • Launched the ”Fancy Items” block on Avito’s homepage, resulting in a 118.4% increase in high-interest item views, from 1 million to 2.18 million views per month. • Developed the ”Like New” recommendation block, which boosted views for similar new items by 86.8%, growing from 800,000 to 1.5 million views per month. • By implementing ”Seller Type” filters, refined search accuracy and increased buyer clicks by 37.3%, reaching 685,000 clicks per month from 500,000. • Implemented in-memory caching in critical sections of system (400k RPM), resulting in a 32% performance improvement for the search filters service

Assortment Matrix Team • Developed and deployed an asynchronous data processing module, adopted by 20 microservices, reducing data import time by 39% (from 126.3s to 76.8s) across the business unit. • Improved query performance of a critical microservice by applying GIN indexes to BSON data, reducing query execution time from 500ms to 300ms, resulting in a 40% improvement. • Reduced application regression by 60% by covering core functionality with e2e tests. • Optimized the CI/CD pipeline by integrating a linter and enforcing mandatory unit/integration tests before deployment. • Minimized service startup time by 19% through the implementation of a parallel Docker image builder and optimizing the number of workers based on performance research.