Document
Portfolio ©2026
Role
AI Product Manager · AI Engineer
Career
21+ Years · Web → Game → Data → AI
Works
10 Works · 100% Hands-on

Geonyeong Kim Geonyeong Kim — AI Product Manager · AI Engineer Portfolio

I don't just bring in AI.
I rebuild how the work gets done.

I built web services, games and platforms for 21 years and started data science in 2017. Now I plan and run AI character chat and content services, and build the models and engines they need myself.

HighlightsNEXUS · NEXON

Key results at NEXUS and NEXON

NEXUS, where I launched a character chat and an AI agent, and NEXON, where I spent 12 years on content platforms and data analytics. Their key results come first.

Certified · 2025.06

MS AI Product Manager

Certified as an AI Product Manager, covering AI product planning, requirement definition and launch management. The same year I spoke at AI Expo on responding to AI tech trends, from LLM to MCP.

  • AI product planning
  • Requirement definition
  • Launch · operations
  • AI Expo 2025 Speaker
2025.08 — NowAI Engineer · AI Tech Lead

NEXUS · NEXUS

Planned, built and launched the character chat ondo and the AI agent ARA

tech references shipped on one shared framework
dialogue LLM I fine-tuned and serve
AI characters in operation
multilingual support by ARA
  • ondo: character chat →

    AI dating app with relationship temperature, daily routines and world engines. I planned and built it, from kickoff on 2026.10.15 to launch in about a month.

  • AI Companion · Stage

    Interactive story where several AI characters talk together on one stage. I planned and built the chat engine and the simulation engine.

  • AI agent ARA

    The companion AI agent of the blockchain game platform CROSS, which I planned and built. Built into the CROSSx app and Chrome extension, it helps users by chat with support, play, trading and asset management. It answers around the clock in many languages. In February 2026 I connected it to Moltbook, a community built for AI agents.

  • Nexus Talk · LiloImage →

    27B LLM for character dialogue (public on Hugging Face) and an image model at $0.02 per image

2011.12 — 2023.11 · 12 yrsPlatform dev · Data science · Business strategy

NEXON · NEXON Korea

Changed how game marketing decisions get made, with data and generative AI

2023 YouTube Works Awards, Best Business Transformation
videos collected and analyzed daily
generative AI platforms (image · voice)
years · platforms and business strategy at a major company
  • Nexon Creators ↗

    Marketing platform connecting games and creators. Led development and business strategy; 2023 YouTube Works Awards winner

  • YouTube · Twitch trend analytics

    Analyzed about 10,000 videos a day by sentiment, qualitative and quantitative signals for strategy and decisions

  • AI voice platform

    Character-specific game voices with Style Speech and Glow TTS

  • AI image platform

    Creating and managing marketing assets, from planning and prototype to a hyper-personalized marketing strategy

How I run AI projects3 Stages

Planning, operations and engineering, connected in one pipeline.

I built pipelines that link planning and PRDs, post-launch operations, events and metrics, and model training so nothing falls between stages. Designers, developers and operators work from the same PRD and metrics.

  1. 01

    Manage

    Plan and agree in a PRD

    I write down the problem and target metrics first, then agree on them with designers and developers in a PRD. Scope, schedule, cost and quality are tracked against the numbers in that PRD.

    Working with AI

    Idea → requirements → PRD → task breakdown → verification runs through an agent pipeline (Davinci). Once the 9-section PRD and per-screen sub-PRDs are out, I look only at the points where a person has to decide.

    accuracy after turning a client task (trade code mapping) into a metric
    gap between cost forecast and real spend
    agent skills used from planning to deploy
  2. 02

    Grow

    Run it and grow it on metrics

    Most of the work starts after launch. I set up work processes that run the same way every day, then use the metrics to decide the next event and what to improve first.

    Working with AI

    Character generation, review and listing on outside platforms run as a daily pipeline. People only reorder priorities based on the numbers.

    AI characters in operation (8,515 video clips)
    listings on 4 outside character platforms
    creator videos analyzed daily at NEXON
  3. 03

    Build

    Finish character chat with tech

    I have trained and served a character chat model myself. So at the planning stage I can tell what will and won't work, and what one turn costs.

    Working with AI

    For a character chat service I LoRA fine-tuned a 27B dialogue model, checked response length, follow-up questions and character breaks with A/B tests, then put it in the service.

    avg. response length (chars) · a human turn is 107
    character breaks or prompt leaks in 10 stress tests
    cost per scene image (was $0.09)
What Build measures becomes the input to the next Manage. I keep this loop running until the project ends.

Selected
Works

7 Case Studies
Made with NEXUS AI Studio · AI character

Latest · 2026.10ondo — Next-Gen Character Chat

Instead of a chatbot that answers right away, a match who lives their own day and texts you first. I designed and built the relationship temperature engine, the daily routine engine, three worlds and a pipeline that turns a photo taken just now into a video. Every match is an AI character, with face, personality, scenes and videos generated in NEXUS AI Studio, the generation engine I built.

A different match on every visit · real world, 2 women · 2 men

A photo taken just now arrives as a video

  • AI characterDaon Lim
  • AI characterIsol Yoon
  • AI characterDaon Lim
  • AI characterIsol Yoon
  • AI characterDaon Lim
  • AI characterIsol Yoon
  • AI characterDaon Lim
  • AI characterIsol Yoon
  • AI characterDaon Lim
  • AI characterIsol Yoon
  • AI characterDaon Lim
  • AI characterIsol Yoon
  • AI characterDaon Lim
  • AI characterSiwon Hong
  • AI characterIan Yoo
  • AI characterSiwon Hong
  • AI characterIan Yoo
  • AI characterSiwon Hong
  • AI characterIan Yoo
  • AI characterSiwon Hong
  • AI characterIan Yoo
  • AI characterSiwon Hong
  • AI characterIan Yoo
  • AI characterSiwon Hong
  • AI characterIan Yoo
What I DesignedCharacter Chat, Next

I designed how the relationship grows

  1. Character01

    NEXUS AI Studio characters

    AI characters generated in NEXUS AI Studio, from face and personality to voice and daily routine. Each match has a job, a neighborhood and a schedule, so they reply later when they're working and in the morning if they were asleep

  2. Relationship02

    Relationship temperature engine

    The temperature rises as you keep talking. At each of the 4 relationship stages, their tone, what they share, how often they text first and when they confess all change

  3. Memory03

    Promises · memory

    They remember what they said, like 'I'll text you later' or 'wake me at 7', and follow through at that time

  4. World04

    Three worlds

    The real, fantasy and historical worlds each have their own tone and messaging medium, and time and seasons pass in each

  5. Media05

    Photo → video

    A photo that fits where they are and how they feel is made with the NEXUS AI Studio pipeline and sent as a 5-second video. Every result is labeled as an AI character

  6. Product06

    Web app · brand

    Chat, threads, shared activities, world travel, PWA install, the ondo.day landing page and launch countdown

Tap a stage in the temperature card to hold the chat on the right at that stage.

9:00 PM
Daon Lim♥ Getting to know
Introduced today
13.9°

Day 1 · 9:00 PM

Hi, I'm Daon. We were introduced today

I'm a pilot, so sometimes I reply late. But I'll always reply

Nice to meet you! Let's take it slow

Message

Every match is an AI character made with NEXUS AI Studio

Three WorldsWith one location chip

Cross into another world
and the tone and medium change with it

Daon Lim · Pilot — Messenger

Want to go to the Han River tonight? They say it's breezy

02AI Framework · IDE2025—26

Davinci

An agent framework that loops from planning to verification on a one-sentence goal

An in-house AI framework and IDE that combines 33 role-based agents with 39 skills. It is the shared development base that lets the whole organization work from planning to deployment on the same pipeline and the same checks. All 9 ONE AI Labs technical references were built and shipped on it.

da:* skills
Specialist agents
Wave parallel speedup
References shipped on it
03Image Model2026

LiloImage

An in-house image pipeline that draws the same character in every scene

A one-line scene a writer types in Korean goes through five stages and becomes a 1664×2216 portrait shot. Everything, prompt rewriting included, runs on one GPU with no external API calls. About 40 seconds and about $0.02 per scene. Every video below was actually used in production.

Per scene
Cost per image
External API calls
Scenes used in production
04Generative Engine2026
ONE AI Labs solutions key visual

NEXUS AI Studio

A character IP engine that turns one seed line into world, scenes, video and chat

A studio engine that builds character IP end to end with an in-house anime engine, a semi-realistic engine and a self-hosted chat model. Every match in ondo, the latest work, was generated here. Eight engines mesh through pipelines, judges, serving and review contracts.

Characters
Designed scenes
Video clips
Cost forecast error
05Language Model2026

Nexus Talk

A 27B Korean model that talks with human turn length and rhythm

Instruct models are trained to 'answer'. Nexus Talk is a 27B model trained on Korean multi-turn character dialogue. It learned how much to say, when to stop and when to hand the turn back. I did the data, training, evaluation and serving myself.

Avg reply length
Length error vs. human turns
Replies closer to human turns
Character breaks · leaks
06Creative Engine2026

StoryMaker

An AI studio that turns a novel into a webtoon with no drawing

Writing, character and world management, storyboard analysis, rendering and speech balloons, all in one app. It turned a 20-episode, 90,000-character web novel into a vertical-scroll webtoon with no drawing work.

20-episode web novel
Auto panels per episode
Hand-drawn images
Production pipeline from manuscript to webtoon
07Asset Engine2026

PerfectPixel

An asset engine that makes 100+ animated sprites from one line

The generative model is only one stage of the pipeline. A prompt compiler, a quality check loop and deterministic pixel post-processing produce sprites that drop straight into a game engine.

Auto-generated motions
Sprite sets
AI providers
Input
Game Lab3 Games · Built & Shipped

One technical question per game.
AI self-play and bot playtests found the answers.

Three games I built and shipped myself: a battle royale FPS, soccer and rope action. All three run right in the browser.

  1. 01

    GOD

    Online FPS battle royale

    48-player battle royale · netcode · 1,000-match bot tests

  2. 02

    KICK·OFF

    11v11 3D football

    Self-play tuning · Decision AI · Voice commentary

  3. 03

    CHASA

    2.5D rope-action adventure

    Rope action · 3D-to-pixel rendering · bot playtests

Game 01Online FPS battle royaleWeb · online match server

GOD

GOD — a Joseon steampunk battle royale

Research Question

“Can I build a 48-player online FPS, match server included, that runs in the browser?”

On the night of the first full moon in 1696, 48 players jump from an airship into a sealed-off Hanyang and fight until one squad is left. I wrote the match server and netcode myself, and bots fill empty slots. Match length and the share of unfair deaths were tuned by running bot-only matches 1,000 at a time. Work started on November 2 and it is now in playtest.

Play nowgod.andrew-ai.app
Live Capture · Lobby · character pick → armory, market, records → deploy → airship drop → landing

Problems solved while building

3 Research Tracks
  1. Netcode01

    30 Hz match server · lag compensation

    The server runs 48-player matches at 30 Hz and rewinds bodies to what the shooter saw before judging hits

  2. Drop02

    Airship drop · kite glide

    Jump off the airship route, open a kite and land anywhere in the city

  3. QA03

    1,000-match bot tests

    Repeated 1,000-match runs tuned match length and the share of unfair deaths

Principles4 Principles · 90 Days

AX means changing how an organization makes things.

I planned and managed AI projects at three organizations: Nexon, Laonpeople and NEXUS. Here are the four principles I kept and what I do in the first 90 days of a new project.

  1. 01

    Turn requirements into numbers

    Before starting, I write down what the client wants as an eval set and success metrics, and we agree on them. Those numbers decide when the work is done. Requirements agreed only in words come back at the end of the schedule.

    → 95%+ HSK trade code mapping accuracy

  2. 02

    Fix scope and cost first

    In an AI project, model call costs matter as much as the schedule. I work out marginal and fixed costs at the design stage and compare them with real spend in production.

    → Cost forecast within 6% · per-image cost $0.09 → $0.02

  3. 03

    All the way to production

    I don't stop at a PoC. I ship models and agents to real users and outside platforms, then run them as pipelines that work every day.

    → 8,515 video clips · listed on 4 outside platforms

  4. 04

    Speak the language of business

    For 21 years I moved between planning, design, development, data and business strategy. With clients I talk in revenue and hours saved. With developers and designers I talk in specs and screens.

    → YouTube Works Awards, Best Business Transformation winner

First 90 Days

  1. Day 0–3001

    Diagnose

    Map the work and mark where AI fits and where it shouldn't go. Set the metrics and eval sets first.

    • Work map · AI fit matrix
    • Success metrics and eval sets
    • Data and security boundaries

    Evidence — Nexon: 2017 data science, video analytics platform designed around what to measure

  2. Day 30–6002

    Shared engine

    Instead of each team buying its own tools, set up one AI dev base and guardrails for everyone. Every later project gets faster on top of it.

    • Shared agent · skill framework
    • Verification gates and cost limits
    • Multi-LLM routing

    Evidence — Davinci: 33 agents · 39 skills, the lab's shared infra

  3. Day 60–9003

    Lighthouse project

    Pick one task on one team and change it all the way. Put it into real operation, then show before and after in numbers.

    • 1 AI workflow in production
    • Before/after report
    • Cost structure dashboard

    Evidence — HSK Assistant: 95%+ trade code mapping accuracy

  4. Day 90+04

    Scale and embed

    Package what worked as skills and copy it to other teams. Document where people decide and where AI takes over, then train on it.

    • Reusable skill library
    • Per-team rollout roadmap
    • Internal training · guides

    Evidence — 1 framework → copied into 9 references

AX Track Record · Organizations I changed2017 → 2026

Seven times, I changed how an organization works.

AX work at Nexon, Laonpeople and NEXUS, from data science in 2017 to our own models in 2026. I kept only the changes where the before and after show up in numbers.

  1. 2017

    01 · NEXON Korea

    Marketing decided by data, not gut feel

    BeforeCreator marketing results judged by gut feel

    AfterStarted data science and built a platform that collects YouTube · Twitch video metadata daily and runs sentiment, qualitative and quantitative analysis

    videos / day collected and analyzed

    → 2023 YouTube Works Awards, Best Business Transformation winner

  2. 2019

    02 · NEXON Korea

    Assets and voices made with gen AI

    BeforePeople made every marketing asset and character voice by hand

    AfterSet a hyper-personalized marketing strategy with an AI image platform and a TTS voice platform based on Style Speech and Glow

    gen AI platforms (image · voice)

    → Planning → prototype → business strategy, hands-on

  3. 2023

    03 · LAONPEOPLE

    From PoC to enterprise product

    BeforeAn organization where LLM adoption stopped at the demo

    AfterHIFENN multi-LLM agent platform (SaaS · on-prem), EZPLANET full-cycle MLOps platform, in-house LLM tuning pipeline

    stage MLOps cycle (prep · preprocess · tune · serve)

    → PO / Tech Lead · LLM Model Research Team Lead

  4. 2025

    04 · LAONPEOPLE

    Real work, run by agents

    BeforePeople looked up a 10-digit trade code for every product name

    AfterCode inference with a Fine-tuned LLM on Korea Customs Service HSK data + Multi-Agent RAG, plus automated tariff, import regulation and origin checks

    HSK trade code mapping accuracy

    → HSK Assistant, a commercial service

  5. 2025

    05 · NEXUS

    An AI agent beside every platform user

    BeforeBlockchain games, wallets and trading were hard for first-time users

    AfterShipped ARA, the companion AI agent for CROSS, in the CROSSx app and Chrome extension. It helps by chat with support, play, trading and assets, and connected it to Moltbook in February 2026

    multilingual user support

    → AI agent ARA

  6. 2025

    06 · NEXUS

    A shared AI engine for the whole lab

    BeforeNew AI and blockchain business unit, no shared dev base

    AfterBuilt Davinci, a shared AI framework with 33 role-based agents and 39 skills, and made every product on top of it

    AI tech references shipped

    → 1 framework → 9 references

  7. 2026

    07 · NEXUS

    Our own models, trained and served

    BeforeGeneral models couldn't hold character consistency or cost

    AfterTrained and served our own: image model LiloImage, 27B chat LLM Nexus Talk, generation engine NEXUS AI Studio

    cost per image (was $0.09)

    → Public on Hugging Face · distributed on 4 outside platforms

Skill
Tree

Nodes
16
Master
9
Branches
5

I started with game websites, moved through platforms, data and enterprise AI, and then built an AI-native lab from scratch. Each period's skills became the base for the next. Click a node to see the evidence.

AI · Model

AI · Models

  1. 2011—2023Lv.4

    Gen AI · TTS

    Since 2019, built an AI image platform and a game character voice platform on Style Speech · Glow.

    Nexon →

  2. 2023—2025Lv.5

    Multi-Agent RAG

    A multi-LLM architecture combining Foundation and in-house LLMs inferred HSK codes at 95%+ accuracy.

    HSK Assistant →

  3. 2025—Lv.5

    LLM fine-tuning · Image models

    I build it all myself: 27B LoRA training, vLLM serving, and an in-house anime image model.

    Nexus Talk →LiloImage →

Agent

Agent engineering

  1. 2023—2025Lv.4

    LLM → MCP

    Spoke at AI Expo on strategy for AI tech trends, "from LLM to MCP".

    AI Expo 2025 →

  2. 2025—Lv.5

    Loop Engineering · Skills

    Designed a shared framework of 33 agents and 39 skills and shipped references on top of it.

    Davinci →

Product

AX · Product

  1. 2004—2011Lv.4

    UI · UX · Graphics lead

    As Gravity's graphics team lead, I designed the UI/UX of the integrated game publishing system.

    Gravity · Softmax →

  2. 2011—2023Lv.5

    Platform planning · Strategy

    Planned Nexon Creators, linking games and creators, and owned its business strategy.

    YouTube Works Awards winner →

  3. 2023—2025Lv.5

    PO · Team leadership

    Led the team as PO and Tech Lead of an enterprise AI product.

    Team Lead, LAONPEOPLE →

  4. 2025—Lv.5

    AX strategy · Cost design

    I design marginal and fixed costs for AI pipelines and keep forecast error within 6%.

    NEXUS AI Studio →

Data

Data

  1. 2011—2023Lv.5

    Data science · Analytics lead

    Started data science in 2017, collecting 10K videos a day for sentiment, qualitative and quantitative analysis. Led the team as Head of Data Analytics at NGINE. TensorFlow Developer · Big Data Specialist L1.

    Nexon 2017— →Head, NGINE →

  2. 2023—2025Lv.4

    MLOps · LLM tuning

    Planned the EZPLANET full-cycle MLOps platform and, as LLM Model Research Team Lead, built an in-house local LLM tuning pipeline.

    LAONPEOPLE →

  3. 2025—Lv.5

    Eval · Metric design

    I prove improvements in numbers with before/after comparisons, eval sets and judges.

    Nexus Talk 39/39 →LiloImage 4.8→7.0 →

Engineering

Services · Infra

  1. 2004—2011Lv.4

    Web dev · Design

    Built and designed the CROSSECTOR web app and the SD Gundam Online game website myself.

    SD Gundam Online →

  2. 2011—2023Lv.4

    Platform development

    Built a creator marketing platform and a voice generation platform.

    Nexon Creators →

  3. 2023—2025Lv.4

    MSA · Multi-tenant SaaS

    Supported SaaS and on-prem at once with FastAPI · Nuxt microservices on Azure.

    HSK Assistant →

  4. 2025—Lv.5

    Multi-runtime products

    I ship desktop and web products myself with Tauri · Rust, Wails · Go and Next.js.

    Davinci →StoryMaker →

Capstone · Lv.MAX

AI PM × Engineer

Hands-on across service planning, PRDs, operating metrics, model training and agent design. I build the pipelines that connect these five areas so the team works from the same standards, and I prove the results in numbers.

About
Me

Geonyeong Kim

Geonyeong Kim · AI Product Manager · AI Engineer

I started in web and game development, then took on platform planning and business strategy. I began data science in 2017 and went on to build generative AI, enterprise LLMs and our own models. Today I plan AI character chat and content services, write the PRDs and grow them on metrics after launch. Where quality is at stake, I fix the models and agents myself.

Career
21 yrs 11 mos
Data science
2017—
Organizations
6
Experience21 yrs 11 mos · 2004.11 → Now
  1. 2025.08 — Now · 1 yr 3 mos

    NEXUS Co., Ltd.

    AI Engineer · AI Tech Lead

    An AI-native lab from scratch

    BeforeNew AI and blockchain business unit, no shared dev base

    After9 tech references shipped on 1 shared AI framework

    • 2025.08 — 2026.07AI Tech Lead · AI service development ↗Led AI services and engines built on AI + blockchain, and oversaw every ONE AI Labs reference
    • 2025.08 —AI agent ARACompanion AI agent for the CROSS platform. Built into the CROSSx app and Chrome extension, 24/7 multilingual support plus help with play, trading and assets; joined Moltbook in 2026.02
    • 2026.01 — 2026.07AI Framework · IDE — Davinci ↗Led an AI framework and AI IDE with 33 role-based agents and 39 skills
    • 2026.02 — 2026.06AI character generation engine · LLM ↗Built an image model for character generation (LiloImage), its pipeline, and a chat-only LLM (Nexus Talk)
    • 2026.10 — 2026.11ondo — character chat (AI dating app) ↗Designed and built the next form of character chat: relationship temperature, daily routines, a world engine, photo-to-video, the web app and the brand (kicked off 2026.10.15, launched in about a month)
  2. 2023.12 — 2025.08 · 1 yr 9 mos

    LAONPEOPLE

    PO / Tech Lead · Team Lead

    Enterprise AX as a product

    BeforePeople looked up and mapped a trade code for every product name

    AfterMulti-Agent RAG infers 10-digit HSK codes, 95%+ accuracy

    • 2023.11 — 2025.07HIFENN — AI Agent Platform ↗Multi-LLM architecture with Foundation + in-house LLMs, LangChain agent workflow engine, FastAPI · Nuxt MSA · Azure, multi-tenant SaaS · on-prem
    • 2025.01 — 2025.07HSK Assistant — trade AI ↗Fine-tuned LLM on Korea Customs Service HSK data, Multi-Agent RAG infers 10-digit codes (95%+), automated tariff, import regulation and origin checks
    • 2023.12 — 2025.06LLM Model Research Team LeadBuilt an in-house local LLM tuning pipeline for agent task automation
    • 2023.12 — 2024.04EZPLANET — MLOps Platform ↗Led planning and design of a full-cycle MLOps platform: dataset prep → preprocessing → tuning → serving
  3. 2022.10 — 2023.07 · 10 mos

    NGINE

    Head of Data Analytics

    Leading a data analytics team

    • 2022.10 — 2023.07Data analytics team leadManaged a range of data analytics projects and led development planning
  4. 2011.12 — 2023.11 · 12 yrs

    NEXON Korea

    Platform dev · Data science · Business strategy

    Changing marketing decisions with data science and gen AI

    BeforeCreator marketing results judged by gut feel

    After10K YouTube · Twitch videos collected daily, decisions made on sentiment, qualitative and quantitative analysis

    • 2017 —Started data scienceMoved into Python and ML analysis of marketing and creator data. All my later AI work started here
    • 2020.02 — 2023.11YouTube · Twitch trend analysis platformCollected metadata for ~10K videos a day and supported strategy and decisions with sentiment, qualitative and quantitative analysis
    • 2019.11 — 2023.11Nexon Creators ↗Platform dev and business strategy for a marketing ecosystem linking games and creators (2023 YouTube Works Awards winner)
    • 2019.11 — 2023.11AI image creation platformFrom planning to prototype of a platform to make, manage and generate marketing assets, with a hyper-personalized marketing strategy
    • 2020.02 — 2023.11AI voice generation platformPlatform dev and business strategy for custom game character voices using TTS models such as Style Speech and Glow
  5. 2006.03 — 2011.11 · 5 yrs 9 mos

    Gravity

    Game dev graphics team lead

    Leading a team that made games

    • 2006.03 — 2011.11Integrated game publishingPlanned the integrated publishing system and led and built its UI/UX
  6. 2004.11 — 2006.03 · 1 yr 4 mos

    Softmax · CROSSECTOR

    Game web · Web application development

    Starting on the web

    • 2005.08 — 2006.03Softmax — SD Gundam OnlineLed game website development and design
    • 2004.11 — 2005.06CROSSECTORLed web application development and set the business strategy
Awards · Talks
  • 2025.05AI Expo Speaker: Responding to AI tech trends, from LLM to MCP
  • 2023.09YouTube Works Awards: Best Business Transformation winner
  • 2020.10Red Dot Award — Brand & Communication Design
Certification
  • 2025.06MS AI Product Manager
  • 2022.02Big Data Specialist, Level 1
  • 2022.02Coding Instructor, Level 1
  • 2021.03TensorFlow Developer Certificate
  • 2003.04Craftsman Computer Graphics Operation
Skills

AI · Models

LLM fine-tuning (LoRA) · Multi-Agent RAG · LangChain · agent workflows · Image generation pipelines · TTS · MLOps planning · vLLM serving

Agent engineering

Loop Engineering · Claude Code · Codex · Gemini CLI · MCP · Agent Skills · Wave DAG orchestration · Eval · guardrail design

AX · Project management · Business

Requirement structuring · eval set design · Scope · schedule · cost management · PO · platform planning · Business strategy · Tech leadership · team building · UI · UX · Figma · JIRA

Data science

Python · ML (2017—) · TensorFlow · Sentiment · qualitative/quantitative analysis · Big data collection pipelines · Metric · eval set design

Services · Infra

Python · FastAPI · TypeScript · React · Next.js · Vue · Nuxt · Go · Rust (Tauri · Wails) · Azure · AWS · Vercel · Multi-tenant SaaS · on-prem

The Making of an AI Product Manager. 21 years and 11 months of work

  1. Intro Prologue: No one becomes an AX engineer overnight. 21 yrs 11 mos, eight pieces. How I became an AI Engineer × AX Engineer
  2. 2004 Web & Game: Web and games taught me about users. CROSSECTOR · Softmax: web applications · SD Gundam Online game website. Gravity: graphics team lead · integrated publishing UI/UX (5 yrs 9 mos) Piece earned — USER: Technology has to be a tool people use.
  3. 2011 Platform: Platforms taught me business. 12 years at NEXON Korea: platform dev · business strategy. Nexon Creators: a marketing ecosystem linking games and creators Piece earned — BUSINESS: Technology has to reach revenue and decisions.
  4. 2017 Data Science: In 2017, I started data science. Analyzed marketing and creator data with Python and ML. 10K YouTube · Twitch videos a day: sentiment, qualitative, quantitative. TensorFlow Developer · Big Data Specialist L1 · Head of Data Analytics at NGINE Piece earned — DATA: Decide with data, not gut feel.
  5. 2019 Generative AI: Gen AI changed how we made things. AI image platform: marketing asset generation · management · hyper-personalization. AI voice platform: game character TTS on Style Speech · Glow Piece earned — GEN AI: A model has value when it solves a business problem.
  6. 2020 Recognition: I learned to prove it with results. 2020 Red Dot Award — Brand & Communication Design. 2023 YouTube Works Awards: Best Business Transformation winner Piece earned — PROOF: Design and transformation are both proven by results.
  7. 2023 Enterprise AX: My first time putting AI into how an organization works. LAONPEOPLE PO / Tech Lead: HIFENN AI Agent Platform. HSK Assistant 95%+ · EZPLANET MLOps · LLM Research Team Lead. AI Expo Speaker · MS AI Product Manager Piece earned — AX: AX is done only when it runs in operations.
  8. 2025 AI Native Lab: I built the way we work into an engine. NEXUS: AI Engineer · AI Tech Lead. Davinci, a shared AI framework: 33 agents · 39 skills Piece earned — ENGINE: Build a shared engine before buying tools.
  9. 2026 Models & Engines: When needed, I build the model too. LiloImage · Nexus Talk · NEXUS AI Studio. 9 tech references shipped · distributed on 4 outside platforms Piece earned — MODEL: Train the model to fit our work.
  10. Now AI Engineer × AX Engineer: Next up: your organization. User · Business · Data · Gen AI · Proof · AX · Engine · Model. Eight pieces linked into one production pipeline
The Making of an AI Product Manager· 0/8
Contact

Let's rebuild
how you work.

I read problems in the data, build the fix with AI, and change how the organization works.
Data analysis, AI development and AX. I own the work from day one to the results.

GY.KIM