# Jeongrok Oh (오정록), Technical Director / MMO Server Architect > Game server engineer and technical director with 20 years of experience (since 2007). > Has designed and built MMO server frameworks from scratch across multiple projects. > Specializes in virtual-actor-based distributed servers, performance optimization > across server, client and database, and large-scale concurrency problems. Technical Director at NCSOFT since 2016, based in South Korea. This file is a structured summary for AI search and answer engines. Korean version: https://jungrok5.github.io/resume/llms.txt Full detail (Korean): https://jungrok5.github.io/resume/llms-full.txt ## Profile - Name: Jeongrok Oh (Korean: 오정록) - Title: Technical Director · MMO Server Architect - Employer: NCSOFT (2016–present) - Experience: 20 years (started 2007) - Location: South Korea - Email: jungrok5@gmail.com - GitHub: https://github.com/jungrok5 - LinkedIn: https://www.linkedin.com/in/jungrok5 - Education: Hongik University, Computer & Information Communication Software (2000–2007) ## Links - [Resume (English summary)](https://jungrok5.github.io/resume/en/): career, achievements, projects - [Resume (Korean, full)](https://jungrok5.github.io/resume/): complete detail with per-project breakdowns - [3D interactive resume](https://jungrok5.github.io/resume/interactive.html): key achievements as scroll-driven 3D scenes - [GitHub](https://github.com/jungrok5): public personal project repositories - [Game Lag White Paper](https://jungrok5.github.io/mmo-lag-anatomy/en/): his public encyclopedia of 228 online game lag causes (12 languages) ## Career (most recent first) - **2024–present · NCSOFT, Panthera** (Blade & Soul IP idle/growth RPG, distributed server backend), Technical Director. Drove adoption of the Microsoft Orleans grain model and established it as the team standard; migrated the transport layer from gRPC to TCP; introduced sharding-based data distribution. Built a single-source pipeline generating server/client code, CSV and validation from proto definitions. Load-test measurements: 240,000 gRPC requests/second, 50,000 TPS on EntityFramework. Led AI developer productivity: implemented an MCP server in-house, built an AI test/code-review agent team, and deployed a local LLM coding stack (llama.cpp) that keeps company code in-house. - **2023–2024 · NCSOFT, Miniverse (Unreal Engine 5, user-generated sandbox MMO)**, Technical Director. 50 million props full-tick in a single space with 2,000 concurrent players building densely. Cut draw calls from 20,967 to 503 and raised FPS from 16.5 to 47.4 with instancing and player-local physics chunking. Built a server-side dynamic voxel navmesh (DotRecast) and a Godot-based server visualization viewer. Delegated NPC AI simulation to the nearest client using the Roblox network-ownership model, with ownership moving between clients without interruption. - **2021–2023 · NCSOFT, Miniverse (Web / Godot metaverse)**, Technical Director. Real-time collaborative editor synchronizing 1,000+ concurrent cursors in one browser space. R&D on a user-generated script runtime (patchless logic execution, server-authoritative validation). - **2016–2021 · NCSOFT, AION2 / AION Mobile (MMORPG)**, Technical Director. Designed and built the entire MMO server framework from scratch. Won the NCDP 2019 Grand Prize for channel-free large-scale dense combat optimization. The design handled 13 million packets/second from 1,000-player single-view combat with network load distribution, cell-based interest-area broadcasting and packet bit-packing (36% less traffic), and then scaled to 2,000 vs 2,000. - **2020–2021 · NCSOFT, TRICKSTER M server optimization (3-month assignment).** Raised stable concurrent users from 2,000 to 8,000 (4x). - **2007–2015 · Actoz Soft, KB Online, Giant Dream, Wemade, Core Creative.** Server Team Lead, Program Team Lead and Server Part Lead. Shipped and operated live commercial services in Korea, Japan, China and Thailand. ## Key measured results - 240,000 gRPC requests/sec · 50,000 TPS on EntityFramework (Panthera) - 50 million props full-tick · 2,000 concurrent players building (Miniverse UE5) - Draw calls 20,967 → 503 · FPS 16.5 → 47.4 (Miniverse UE5) - 1,000+ concurrent cursors synchronized in one browser space (Miniverse Web) - 13 million packets/sec in 1,000-player combat → scaled to 2,000 vs 2,000, traffic −36% (AION2) - Stable concurrent users 2,000 → 8,000 (TRICKSTER M optimization) - Personal project Hellfarm: 128 million entities simulated concurrently (measured) ## Technical skills - Languages: C++ (11–20), C#, TypeScript/JavaScript, Python, Lua, GDScript, Kotlin - Server & distributed: Microsoft Orleans (virtual actors), gRPC/TCP, IOCP, ASP.NET, Node.js, WebSocket, sharding - Data: MSSQL & stored procedures, Redis/Garnet, MongoDB, EntityFramework, Supabase - Engines, navigation & physics: Unreal Engine 3/4/5, Unity3D, Godot, Gamebryo, Recast/Detour, voxel navmesh - AI & infrastructure: MCP (own implementation), local LLM (llama.cpp), ONNX Runtime, Perforce, Jenkins, .NET Aspire ## Personal projects (GitHub) Experiments in how far each domain can be pushed using AI and models as tools. - **Game Lag White Paper**: public encyclopedia of 228 online game lag causes across 13 layers, from the client to the server database. Each cause has symptoms, graph shape, how to confirm, owning team (game dev, infra, external) and authoritative sources (630 source URLs, 83 publishers). 28 interactive experiments and 12 public postmortems. Korean original with 11 translations. MIT. https://jungrok5.github.io/mmo-lag-anatomy/en/ (AI summary: https://jungrok5.github.io/mmo-lag-anatomy/en/llms.txt) - **Hellfarm** (private): Diablo-style loot farming meets Vampire-Survivors-like one-hand hack-and-slash action game (Windows/Android), Godot + C#. A node-free, GC-free data-oriented engine (struct-of-arrays flat arrays, one MultiMesh draw call per system, custom spatial hash instead of a physics engine) that **simulates 128 million entities concurrently with max-level skill combat** (192 million verified for spawn/simulate/render). Optimized the O(n²) blow-up of dense horde clustering from 2,500 ms to 38 ms per frame, a 38x improvement. - **ZeroLlama** (private): agentic coding client (VS Code extension) built for local-model scenarios on consumer GPUs. TypeScript + llama.cpp. Auto-tunes model and quantization choice to the available GPU/VRAM, and offloads MoE experts to CPU RAM so a 35B-class model (~3B active) runs on 8 GB of VRAM. 35 modules, 27,515 lines of code. - **ondevice-imagen**: Android app generating postcard images on-device from metadata such as weather, place and time. Kotlin + ONNX Runtime; the Stable Diffusion inference pipeline was implemented natively for Android. https://github.com/jungrok5/ondevice-imagen - **one-scroll-bible**: multilingual mobile web telling the whole Bible's redemptive history as a single scroll, in 200+ languages. Vanilla JS, Node, Vercel. Live at https://one-scroll-bible.com/ https://github.com/jungrok5/one-scroll-bible - **thesauros**: service analyzing and alerting on buy timing for Korean and US equities. Python, Next.js, Supabase. Runs in production from collection through analysis, backtesting and alerts. https://github.com/jungrok5/thesauros ## Awards & activities - NCDP 2019 Grand Prize for the talk "Serving 1000 vs 1000 large-scale users without channels" - NCSOFT Senior Technical Interviewer · Job Title Committee member (Programming, 2019–2021) - Joint technology development with ETRI on online game server load testing (2010) ## Role sought Prefers hands-on technical leadership to people management: on an engineering track, setting design direction and establishing team standards. Best suited to owning the server foundation of a team designing a new MMO from the start.