Career Description

Hi, I’m Seungha Lee — a backend engineer who enjoys nothing more than the time spent solving problems.

Seven years in software and engineering — Java backend (2 years) → AI/GPU infrastructure R&D (about 2 years) → Go backend (3 years).

In my current role I work on payments, notifications, transaction history and data pipelines, mainly designing for data consistency in distributed systems (pre-emption, idempotency, recovery), and I also handle CI/CD, Kubernetes manifests, the gateway and DB DDL for my services. I design the failure path first and turn review feedback into written team principles. I’m applying for both backend and DevOps roles.

Skills

Work Experience

NEXON UNIVERSE — MapleStory Universe

Backend Developer · Platform Development Team · Nov 2023 – Nov 2026 (3 years)

NEXON UNIVERSE (founded 2022, a Nexon affiliate) builds MapleStory Universe (MSU), a Web3 platform that brings blockchain to the MapleStory IP. I developed its backend on the platform development team.

I started on the on-chain/off-chain synchronization layer and went on to work on transaction history, bookmarks, the marketplace BFF, account/payment, unified notifications, a policy service, a CDC data pipeline and an operations AI bot.

Go gRPC Kafka PostgreSQL Redis MongoDB Snowflake AWS EKS Helm Vertex AI

My core role: keeping data consistent in distributed systems — preemption · idempotency · CDC.

Portfolio (KR) — problem · approach · result → Design Decisions (KR) →

What I did

1 On-chain / off-chain synchronization (Data Engine)

Main work Nov 2023 – Apr 2024

GoSagaRedis distributed lockHelm

• Added input validation, request/response conversion with unit tests, and gRPC error handling while wiring a new core service into the game-facing gRPC service (business-provider)
• Added currency/character validators for the Saga v2 migration and reworked NFT item validation from per-item state lookups to a single list query
• Fixed nickname duplication at character mint: dropped the existence check, preempted with a Redis lock, released only on successful mint — a small fix, but a pattern I kept reusing
• Ran Helm rollouts for new services, reproduced an incident with throwaway deploys and rollbacks, and deployed to dev first before test/qa promotion

2 Transaction history · bookmarks · marketplace BFF

Main work Jun 2024 – Sep 2025 · maintained afterwards

gRPCKafkaPostgreSQL advisory lockRedis NXCDCSnowflake

• High-volume transaction ledger (api-activity)
  • Reorganized layers, generalized the consumer and moved to message format v2, then extended it to trades, mints, transfers, Swap & Warp, bridge-in and USDT
  • Promoted the transaction hash to a unique key and routed malformed events to a DLQ
  • Protected concurrent ingestion with Redis NX + PostgreSQL advisory locks
• Query performance — OR → UNION ALL rewrite with the DBA, cursor-based infinite scroll, a conditional-caching option added to the shared cache decorator
• Emitted notification events when a transaction result is finalized; moved Snowflake ingestion from Kafka publishes to CDC, removing the publish-failure loss path
• Bookmark / view-count service (api-bookmark) — core features: idempotent writes, Redis counters moved to Kafka-driven aggregation, daily/total limits, account-deletion hook
• Marketplace web BFF — moved bookmark/activity calls to the new services, implemented bookmark and on-sale listings in the inventory service, parallelized rendering. Traced intermittent pod restarts to gRPC calls inside async loops holding resources and fixed them with timeouts and chunked calls
• Swap & Warp cross-chain transfer history and queries (incl. bff-swapnwarp)
• Added bookmark-event ingestion to a Java/Spring Kafka consumer; fixed bugs in the marketplace BFF and game-data service before launch

3 Account · payment · unified notifications

Main work Jul 2025 – Aug 2026

WorldpayRedis LuaSSEgRPC streamingMongoDB

• Worldpay (FraudSight 3DS) card verification & re-verification
  • Mapped vendor 4xx to 200 + outcome codes and 5xx to errors
  • Made the daily attempt counter atomic with a Redis Lua script
  • Encrypted part of the card data and added blocking of abnormal attempts per payment-method combination
  • Added a re-verification path that confirms identity by card when OTP is reset
• Stablecoin (USDT) launch — extended the off-chain pre-deduction/rollback service for USDT (per-token limit tables, pre-deduction and row locking generalized per token), replaced the rollback topic with a dependency on transaction result topics, and added wallet-address validation to the blockchain supporter and BFF
• Unified notifications (SSE) — absorbed JWT verification to drop the account-service dependency, generated URLs server-side, and split bulk recipients into a gRPC streaming RPC (500-row chunks, 50,000 cap)
• Full-user data migration batch for the KYC vendor change (Argos → Sumsub) — range-partitioned parallel runs, resumable cursor, call-quota management; completed in production
• Discord auto-badge and role-sync batch; extended the community-notice crawler with new boards and resumable failures
• 24-hour lock on security-option changes after an e-mail change — moved from Redis TTL to a persisted column, with primary reads to close the replication-lag window

4 Policy service · data pipeline · ops automation

Main work Jan 2026 – Nov 2026

DebeziumSnowflake Stream/TaskVertex AIOpenSearch

• Policy service (api-sentinel) — ACL (wallet/IP allow & deny lists), rate limiting (sliding + fixed window with jitter), lockout after repeated OTP failures (keep/reset failure counts by configurable policy), runtime feature flags
• PostgreSQL CDC → Snowflake — Debezium connectors, Stream/Task, a replication-slot keepalive job, conditional tasks to cut cost
• Error-triage bot — replies in-thread to Slack error alerts with an LLM first-pass analysis (GitLab commit diffing, OpenSearch history, escalation); handled a vendor model deprecation with model swap, retry/backoff and a semaphore

5 Infrastructure · CI/CD · gateway

Throughout

GitLab CIDockerHelm · EKSRundeckEnvoy

• CI/CD & infrastructure — operated GitLab CI, Dockerfiles, Helm/K8s manifests and DDL for the services I owned; new services shipped with their full deployment set
• Envoy gateway — added and tidied gRPC → REST routes and auth settings whenever services were added; moved routing/port settings of my services (notifications, blockchain supporter) as part of the team-wide gateway gRPC → HTTP switch
• Reviewed all externally exposed APIs (Swap & Warp, notifications, BFF) and added authentication (team of three)
• Led the split of the company-wide proto repository into domain modules

6 Back office · operations tools · other

Mar 2025 – Oct 2026

• Back-office item/token sanctions (single, batch, bulk) and a sanction-history view
• Operations console — Discord message re-send, permission-domain split, wallet ↔ SNS lookups
• Item-probability disclosure service: data storage and versioning

Tech Stack

UCLICK — AI Engineering R&D

AI Engineering R&D · AI Service Team 1 · Jan 2022 – Oct 2023 (1 yr 10 mo)

• NVIDIA Omniverse R&D (Jun 2022 – Oct 2023) — robot assembly and simulation in IsaacSim with ROS messaging (Python); VR device integration in CreateXR with an UnrealEngine4 single-player sample (C++)
• GC-GPU solution incident diagnosis (Apr – May 2022) — kernel-panic log analysis and root-cause fixes at the environment and source level
• NVIDIA GPU MIG support (Mar – Apr 2022) — on-site MIG enablement for a Korean financial client’s dev/prod servers; shell script that recreates MIG profiles after reboot
• GPU monitoring with NVIDIA deepops (Jan – Feb 2022, sole owner) — Prometheus/Grafana via Ansible; repackaged as scripts and images for air-gapped networks

UCLICK — Backend Developer

Backend Developer · R&D Lab 2 · Nov 2019 – Dec 2021 (2 years)

• UNO-GATEWAY (Sep 2020 – Dec 2021) — dynamic OPEN API platform (validation, transformation, auth, rate limiting, routing); WebSocket + performance view, connectors for 7 databases and Elasticsearch/Hazelcast/Redis, Ribbon load-balancing tests, packaging
• UNO-GATEWAY MDB project for a Korean securities client (Sep – Oct 2021, sole owner) — JSON ↔ fixed-format message conversion over TCP
• UNO-MSP (Feb 2020 – Dec 2021, lead developer) — Oracle Cloud MSP platform across four releases: real-time resource collection, dashboards, billing integrity, OCI alarm/rule sync, Slack/mail alerts, monthly reports, HA agents, Zendesk integration, JDBC DB monitoring
• LISTEN2ME-SURVEY (Sep – Oct 2020) — spreadsheet-to-survey web app; API contract and statistics dashboard data

How I Work

Design

• Failure paths first. Pre-deduction/rollback, DLQ and retries, CDC — I decide how to recover before I implement. Defect fixes are merged separately from the feature, so they survive if the feature is rolled back.
• Locks and idempotency chosen per event type. Requests and results that arrive out of order get look-up-then-create/update under an advisory lock; single-message events get Redis NX; unique keys are the last line of defence underneath.
• Policy as configuration, not code branches. A policy change should change a setting, not a commit — and settings are injected at boot, never read on the request path.
• Services don’t write outward themselves. Analytics and downstream loads flow through CDC, which removes “the app failed to publish” as a loss path altogether.
• Complex code is split behind interfaces to keep coupling low and layer boundaries intact.

Operations

• Deploys, manifests and DDL for my services are my job. GitLab CI → Helm/Kubernetes → Rundeck, a different deploy path per environment, and DDL that goes through DBA review — one person sees it through. New services get wired into the pipeline by me.
• A single request must be traceable end to end. I design the correlation keys first, then verify after each deploy that Datadog, Jaeger and OpenSearch actually follow the request.
• Debugging starts from a hypothesis. Narrow the scope by environment and context, then confirm with logs. Repeated ops work gets automated, and after a recurrence the closing criteria gain a new item.

Collaboration

• Separate “current assumptions” from “needs verification” when working with DBA, data engineering, security, product, front-end and blockchain teams.
• Write down design intent and the reasons behind decisions. Diagrams first, open to a better direction; the Design Decisions page on this site is that record.
• Review feedback becomes team principle. Payment and security features go through a second review round, and rules like “no log-and-swallow on external calls” and “inject config at boot” are documented and carried into the next piece of work.

Other Experiences

Certifications

Korea National Open University — Computer Science

Mar 2021 – Feb 2024

Enrolled to ground my engineering in computer science fundamentals and took major courses only — operating systems, compilers, data structures, algorithms, databases, C/C++/Java, security and cloud computing — with A+ in each of these courses.

Korea Polytechnic — Data Convergence Software

Mar 2019 – Nov 2019

Hands-on Java/Spring, Python, R, Linux and database training, including TensorFlow regression, large-scale customer SQL analysis with Oracle, and JSP/Open-API web apps.

Ewha Womans University — Mathematics Education

Mar 2011 – Feb 2016

Mathematics major (calculus, linear algebra, geometry, probability and statistics, analysis, topology, number theory, abstract algebra) — trained to solve problems through definitions and proofs.