Highlights from recent roles, then other workplace projects and the archive.
Highlights
·NAVER Cloud
MLX Data Manager
A data management system designed for machine learning workloads, offering a Hugging Face-compatible interface for ease of adoption, along with data version control and lineage tracking.
Ingested and aggregated operational data—from Kafka into S3 (ORC format)—for analytics, ensuring integrity with deduplication and exactly-once delivery semantics.
Flow (Inter-domain logistics state management system)Handling write-heavy workloads with Cassandra and message ordering guarantee with a distributed queue.javacassandrakafka
·Coupang
Flow (Inter-domain logistics state management system)
Handling write-heavy workloads with Cassandra and message ordering guarantee with a distributed queue.
Ecosystem Pipeline (Durango)Distributed pipeline so the simulator runs in bounded time as the world grows: Dockerized jobs, spatial chunking, and "no audience, no play" cost control.distributedawsscalabilitygame
Distributed pipeline so the simulator runs in bounded time as the world grows: Dockerized jobs, spatial chunking, and "no audience, no play" cost control.
Ecosystem Simulator (Durango)Procedural world vegetation driven by soil, climate, and biomass rules—scaling heavy simulation with precalculation and GPU-friendly workloads.c#openclsimulationgame
Durango — server & overviewDistributed MMORPG server at Nexon—overview of the system and how ecosystem sim, pipeline, and related pieces fit together.gamepythondistributed
A Python library to encode and decode any arbitrary data in base62 (duosexagesimal; using 0-9, A-Z, and a-z) for URL-safety. I was recently able to convince my colleagues to use this library for work.
Contributing compute to BOINC—on the order of 127 GFLOPS on average for climate modeling, prime search, stellar-stream simulations, and similar projects.
Curated a collection of interview questions and solutions from various tech companies, for personal reference and to support fellow engineers preparing for interviews.
Upstream work on Pinterest Secor (Kafka to cloud storage), including handling map types in message serialization. Pull requests: #982, #965, and other authored PRs.
An ambitious project to solve urban traffic congestion problems. My primary responsibility in the team is to build a scalable system with Amazon EC2 platform, and to build in-house software for data analysis.
The project radioactivity had started in March of 2011 in an attempt to help the general public to be alerted regarding the spread of radiation and other radioactive materials originated from the catastrophic incident in the Fukushima nuclear power plant in Japan, by providing visual information of city-wise radition levels.