2025.08 ~ Present
S2W
SAIP, an ontology platform
Primary frontend engineer on SAIP · Product planning (1 of 3) · On-site adoption (FDE)
Gathered requirements on site · Designed the ontology schema
2 PoCs · 1 full engagement S2W only
Taeho Kim 김태호
Frontend · FDE
domybestatleastonceaweek@gmail.com · linkedin.com/in/taeho-kim-0a504a335 · taeho.world/en
I gather requirements at customer sites, design the ontology, and build the screens people actually use.
At S2W, I'm responsible for the frontend of SAIP, an ontology platform, and I'm one of three people planning the product.1 I direct and verify; AI agents implement.2
Business analytics time cut by 95%+
95%+
Cut in business analytics time
Built SAIP's 26 frontend screens solo over 8 months
26 screens
SAIP frontend, built solo over 8 months
Ontology schema design: ~40 entity types · ~70 relations
~40 · ~70
Ontology schema design: entity types · relations
Read the case studydomybestatleastonceaweek@gmail.com The address is selected. Please copy it.LinkedIn (opens in a new tab)
- Now
- S2W / SAIP2025.08 ~ Present
- Career
- Frontend since 2020.026+ years · Frontend engineer at 4 companies
- FDE
- Since 2025.08Expanding from frontend into FDE
- In the field
- 3 PoCs · 5 demos · 3 full engagementsS2W and TmaxCoreAI combined · Led, with colleagues
- As of
How I work: Problem framing · Product design · Building · On-site adoption
Summary
Whenever good UI alone didn't solve the problem, I widened my scope.
Today that includes ontology design, customer data analysis, and gathering requirements on site.
Customer site Hands-on and operations staff
- 1 Listen on site Gathering requirements firsthand · Finding where the data lives and who owns it Done solo
- 2 Structure it Ontology schema design · Designing how the data joins Done solo
- 3 Take it to planning Feature proposals and specs · 1 of 3 on product planning Led
S2W · SAIP Product · Features added to the plan
- 4 Build the screens SAIP frontend · 26 screens · 8 months · as of 2026.08 Done solo
- 5 Put it to use Using SAIP directly in customers' business analytics · Iterating until the goal is met Led · In real use since 2026.07
Back to 1: feedback · new requests
Outside the loop · Closed-network install · separate team
I hand off the frontend build.
What I do · Done solo and Led are defined in chapter 4Another team's work
Case study
SAIP · Ontology platform · S2W · 2025.08 ~ Present · My time on it
Business analytics time cut by 95%+
Every month, the customer's analysts gathered performance figures and, when a number looked off, asked the team behind it for the cause before writing the report. But it was hard for them to judge whether the explanations held up.
My part: on-site requirements gathering, ontology schema design, customer data analysis, and frontend development as the primary engineer. I led product planning (one of three planners) and on-site adoption together with colleagues. Colleagues led the servers and AI agents, and another team handles the closed-network installation.
Case contents
- 1.1 The problem on site
- 1.2 Listening on site 100 reports → the pattern from 8~10
- 1.3 Structuring the data 7 axes · 700,000~1.25 million rows a month → about 200 records a month
- 1.4 Verification 20 or more records hand-checked against the originals · no mismatches
- 1.5 Building the screens Data computed on request · schemas checked in real time
- 1.6 Putting it to use · Result Reporting-ready dashboard and Q&A chat · in real use since 2026.07
- 1.7 Taking it to planning Proposed what to take out · built the new Overview screen
Building with AI
I structure problems, break down the work, and define context and criteria.
I design environments where AI agents can work reliably, then review the output and fix what fails.
- Me 1 Problem framing Structural and experience decisions
- Me 2 Direction Designing an environment where agents can work reliably
- Agent Writing code My recent code, this site included
- Me 3 Verification The high-stakes paths first I fix what fails and go back to Direction
- Me 4 Responsibility For the outcome, from problem definition through implementation and verification
Evidence · This site's git history
Commit history of this site's public repository. In 2022.06~2025.11, 0 of 107 commits carry the AI co-author trailer; in 2026.03~2026.09, 38 of 42 do. Measured 2026.09.30, merge commits excluded.
AI agents wrote my recent code, including this site. I'm responsible for the problem definition, the structural and experience decisions, the direction, the verification, and the outcome.
A choice I made
I chose to build with AI instead of typing everything myself. It lets me cover more ground, faster, but it takes discipline to read and verify the result.
Career
The common thread: I took on the screens and saw them through until people used them.
Previously, I worked on AR try-on, chatbots, academic search, smart-city and video monitoring, and labeling tools.
Segments of the career route
- Frontend at deepixel: sole frontend engineer on StyleAR · 2020.02~2021.06
- Frontend at TmaxCoreAI: frontend engineer · 2021.07~2023.03
- Frontend at Innodep: video monitoring and web VMS · sole FE on the labeling tool · 2023.04~2025.08
- Frontend at S2W: primary frontend engineer on SAIP · 2025.08~Present
- Planning at TmaxCoreAI: team lead and PM from month 3 · 2021.07~2022.12
- Planning at Innodep: proposed the labeling-tool idea · planned the canvas labeling editor · 2024.10~2025.08
- Planning at S2W: 1 of 3 on product planning · led feature proposals and specs · 2025.08~Present
- Data and ontology at S2W: ontology schema design · customer data analysis · 2025.08~Present
- Customer-facing at TmaxCoreAI: drawing out, framing, and pinning down the real requirements for RISS · 2021.12~2022.12
- Customer-facing at S2W: going on site to gather requirements firsthand · 2025.08~Present
2023.04 ~ 2025.08
Innodep
Video monitoring and web VMS
Frontend engineer · Sole FE on the labeling tool
Labeling time cut by 80%+ Actual use by in-house labelers · during my time there
React · Zustand · TanStack Query · Tailwind CSS · HTML Canvas · Styled-Components · WebRTC · Angular · SCSS · OpenLayers · Recoil
4 projects · VUNex MLOps · VUNex AI · Smart City · AI Camera Show projects
VUNex MLOps
- Proposed annotating directly on the video, the idea the product grew from
- Planned, designed, and built the canvas-based labeling editor
- Label an object once and it propagates across every frame (a 10-second clip at 30fps is 300 frames, each boxed by hand in traditional tools)
- Labeling time cut by 80%+ Actual use by in-house labelers · during my time there
- In one month, six labelers worked through 1,000+ hours of footage Footage length, not working hours · during my time there
- The hardest part was data integrity: keeping the video object and the label objects in exact frame-to-frame sync, so a propagated box lands precisely where it should across all 300 frames
React · Zustand · TanStack Query · Tailwind CSS · HTML Canvas
VUNex AI
A product that detects intrusion, falls, fire, and abnormal behavior on factory, retail, and office cameras
- Built the canvas layer that draws detection boxes over live video (solo)
- Co-built the frontend side of WebRTC streaming (1 of 2)
- The hardest part was locking the bounding-box render cadence to many simultaneous live streams, so boxes track their objects without lag or jitter as channels scale up
Up to 25 channels per machine · 16 streamed at once · Latency of 300 ms or less · Adopted by 6 customers (3 in Korea, 3 abroad) Product specs · during my time there (to 2025.08)
React · Zustand · TanStack Query · Styled-Components · Tailwind CSS · WebRTC · HTML Canvas
Smart City Total Management System
- Owned 2 custom screens · Contributed 30% of a third screen
- On those screens my focus was performance: drawing very large numbers of items on the map while keeping render load low, so the browser stays responsive as the count climbs into the tens of thousands
Used in 5 municipalities · 1,500~15,000+ cameras per municipality · Up to 9 control views at once The platform's scale · during my time there
Angular · SCSS · OpenLayers · React · Recoil · Styled-Components
AI Camera
Each camera is configured for sensitivity, regions of interest (ROI), and schedule, so detection runs on site without a central server
React · Recoil · Styled-Components
2021.07 ~ 2023.03
TmaxCoreAI
Chatbot · academic search
Frontend engineer · Team lead and PM from month 3
HyperChatbot adopted by 2 customers (a university and the Korean Army) While I was there
React · Redux · Styled-Components · Recoil
2 projects · HyperChatbot · RISS Show projects
HyperChatbot
- Worked across the chatbot's screens and modules
- Designed and implemented the flowchart-editing UX for non-developers
- The hardest stretch was the flowchart library: its learning curve, plus a mid-project major-version bump that forced a large rework of the data architecture beneath it
The builder is a visual flowchart editor that allows up to 1,000 nodes in a single flow Product spec · while I was there
React · Redux · Styled-Components
Research Information Sharing Service (RISS)
Korea's national academic-research information service, opening university research to the public
- As development PM, took part in planning every screen instead of writing code, set the direction, and divided the work among three developers
- The redesign I worked on is still live
The hardest part wasn't technical. It was requirements discovery: the client often didn't know what they actually wanted, so much of the job was drawing out, framing, and pinning down the real requirements.
React · Recoil · Styled-Components
2020.02 ~ 2021.06
deepixel
AR try-on · first job
Sole frontend engineer for my entire tenure · AR try-on widget Done solo
About 200 customers · About 3,000 products · 1M+ daily visitors on the customer pages carrying the widget · 10,000~30,000 daily API requests During my tenure (2020.02~2021.06)
Vue2 · Vuex · HTML Canvas · WebGL · jQuery · Node.js · Express · MySQL
1 project · StyleAR Show projects
StyleAR
Real-time AR virtual try-on for jewelry, fashion, and beauty, placed on the store's own page as a floating button
- Widget embedded on customer pages · Weekly widget releases · About 50+ in total During my tenure
- Focused on reducing bundle size
- The hardest part was bundle size: the widget loaded on top of each customer's page, so the package had to stay small enough that it never slowed the host site down
- Simple Node CRUD · Contributed to the overall service design
Vue2 · Vuex · HTML Canvas · WebGL · jQuery · Node.js · Express · MySQL
Levels
Only what I've actually done. I left the gaps in, at the bottom.
| Done solo Owned start to finish | |
|---|---|
| Frontend development | StyleAR (entire tenure) · VUNex MLOps labeling tool · 26 SAIP screens in 8 monthsAs of 2026.08 |
| Requirements gathering | 2 companies · Met customers in person · Tracked down the data and its owners · Set up follow-ups myself |
| Ontology schema design | ~40 entity types · ~70 relations |
| Customer data analysis | Data from 500+ stores · Designed how the data joins · Reviewed potential insights with AI agents |
| Led Drove it, with colleagues | |
| Product planning | 1 of 3 · On SAIP since 2025.08: planning, design, and development |
| On-site adoption | 3 PoCs · 5 demos · 3 full engagements (11 total)S2W and TmaxCoreAI combined · Led, with colleagues |
| Customer feedback loop | Iterated until the goal was met · See the result in chapter 1 In real use since 2026.07 |
| Incident response | Took the call · Aligned with sales · Fixed it myself |
| Contributed With someone else leading | |
| Backend · API | CRUD · Simple API work · No experience designing or configuring servers |
| Not yet No experience yet | |
| Deployment · Infra | 0 No server, infra, or closed-network install experience yet · Today I hand off the frontend build · Closed-network install is the next step |
Working with me
I'm best at structuring ambiguous problems and building the first working version.
I work best taking on the initial structure and execution, alongside teammates who excel at optimization and long-term operations.
- 1
I start with the reason and context, not the feature.
Before building a feature, I ask why it's needed and how it will fit naturally into the existing flow.
- 2
Even when AI and automation step in, users should be able to see what's happening, decide for themselves, and undo it.
Evidence · I built the UI for the Agent Panel, the AI assistant inside SAIP's editors: it drafts proposals from natural-language requests, and the user reviews each one, then approves or rejects it.
- 3
Builders should absorb complexity and leave users with the shortest, clearest path.
I reduce the rules users have to learn.
- 4
I look after both the product and how it's used on site.
Evidence · I went on site to gather requirements firsthand, and I used SAIP directly in customers' business analytics workflows.
- 5
What I take on, I see through to the end.
I stay with an idea until it becomes usable. I define the problem, shape the structure and experience, learn whatever technology is needed, and look after it until it's done.
- 6
A choice I made · I chose breadth over depth.
I can take a product from planning through the UI to on-site adoption, but I'm not a specialist in every area.
Off the clock
Random Korea Travel
A travel roulette that picks a destination in Korea for you once you set a starting point and a travel time.
Visit the site Random Korea Travel (opens in a new tab)
Tetrapolis
A 40-question test that maps your politics on four axes, not a single left-right line.
Visit the site Tetrapolis (opens in a new tab)
Once a game hooks me, I tend to go deep.
| Game | Record | How I play |
|---|---|---|
| PUBG | ASIA #120 | AUG + Mk12. Survival and positioning before taking the fight. |
| Overwatch | GRANDMASTER | Reinhardt, shield up, holding the line. |
| Lost Ark | DESTROYER #170 | Hammer in hand, taking every raid head-on. |
| Teamfight Tactics | MASTER | Min-maxing every round. |
| Football Manager | 1,000H+ | Countless seasons with Manchester United. |
| Personal bests · not current ranks · Football Manager: total hours played | ||
Contact
Happy to chat.
domybestatleastonceaweek@
Your browser's print dialog can save it as a PDF.
Online: taeho.world/en · As of