Heeguk Son

Backend Software Engineer on NAVER Shopping, building high-traffic distributed services in Java and Spring Boot with Kafka, Redis, and Elasticsearch. Also built multiple LLM-powered tools for code review, operational automation, and engineering productivity.

Key work:

(1) Built an LLM-based production log triage agent that runs daily in CI, analyzing warnings and errors from 10+ services against a version-controlled knowledge base of prior team judgments; filtered recurring benign patterns, surfaced potential production risks, generated daily issue reports, and auto-filed Jira tickets for critical findings, with operator feedback incorporated through reviewed pull requests.

(2) Engineered dual-cluster failover and migration to a multi-region triple-cluster architecture spanning application servers, Redis, Kafka, and Elasticsearch; validated 100K+ API RPS under simulated failure of two clusters through replication-lag analysis and failover testing.

(3) Designed a Kafka-based event-driven replication pipeline maintaining eventually consistent read models for API serving; routed real-time and resync events through separate processing paths within the same consumer flow, sustaining 10K combined events/sec during backfills of billions of product events without increasing real-time processing lag.

(4) Developed CI validation workflows with Java-based static checks to detect cache misuse, API compatibility breaks, and dependency drift, running automatically in GitHub Actions; surfaced findings in PRs, with zero cache-related production incidents or API compatibility regressions since 2023.

(5) Eliminated duplicate approval processing with annotation-based Redis distributed locking; load-tested pub/sub vs. spin-lock at 10K RPS and selected spin-lock for short lock-hold times and bursty workloads with no measurable latency impact.

(6) Reduced Elasticsearch aggregation time for 600K+ stores from 1 hour to 5 minutes using eager_global_ordinals and query restructuring, enabling store-level search features.

(7) Automated code review with an LLM-based PR review workflow that evaluates Git diffs against coding standards, architecture patterns, and defect history, catching logic and dependency issues before human review and cutting manual review comments by roughly 40%.

(8) Implemented an automated spec synchronization pipeline that detected changes to source-of-truth Wiki requirements and opened pull requests to update version-controlled implementation specs, preventing specification drift during feature development.

Software Engineer on trade finance and foreign exchange systems at one of South Korea's largest commercial banks.

Key work:

(1) Refactored legacy batch logic and optimized SQL across 10+ export/import trade-finance batch jobs written in C, improving reliability and ensuring on-time completion of overnight processing.

(2) Led end-to-end integration testing for trade-finance settlement workflows with Samsung Electronics. Zero production incidents were caused by integration issues over two years of releases.