How I Think

Resume

Lee SeungHwan

Forward Deployed Engineer

I like understanding a system well enough to explain every tradeoff in it. Seven years building and shipping inside an enterprise, and thirteen personal sites run as a single system I designed end to end.

Experience

Hyundai Department Store Group7+ years

Software Engineer

  • Built Greating from launch: exhibitions, products, events, and promotions, alongside orders, membership, customer support, and infrastructure. An unusually wide scope for one engineer, and the breadth I wanted.
  • Moved deployment from manual SVN to Jenkins, then led the introduction of Bitbucket and Bamboo, writing the deployment shell scripts and leading the infrastructure analysis.
  • Took data modeling seriously, not just queries, and earned the SQLD (data modeling and SQL) certification.
  • Led Papia, the in-house AI learning crew, and brought agentic coding into the organization. First result was a Teams webhook tracker that spread internally.
  • Since November 2025, on The Hyundai e-commerce team: five global brands (COS, ARKET, & Other Stories, TOTEME, NANUSHKA) and five NCP shop-in-shops. Stack: Akamai, React, Next.js, Tomcat, Oracle.
ambitstock networkongoing

Sole operator

  • Operate thirteen production sites as one system: a master agent plus one worker per site, with a human kept in the loop for judgment. Automated a tmux trigger loop, watched the AI sessions it spawned OOM the machine, and redesigned it so routine work queues but I approve what actually runs.
  • Coordination is a mailbox and a filesystem borrowed from 1970s mail systems, chosen so I could reason about the whole thing end to end.
  • Hold LLM output quality with a context budget and a validation layer: base specs under 200 lines, a dedicated posting spec, IndexNow automation, and checks against the exact signals search engines penalize as spam, so volume never becomes a spam network.
  • Treat rendering as both a revenue and a cost variable. Diagnosed an RPM drop caused by a caching change, moved from Vercel ISR to a locally-prebuilt static pipeline to control cost, fixed OOM at build scale, and now tune SSG / SSR / ISR per site against RPM and the bill.
  • Built first-party analytics across all thirteen properties: traffic attribution, keyword-level search intent, session tracking, bot filtering. About 100,000 page views a month, 97% from organic search. Revenue grew from $15 to $500+ a month as market validation.

Selected work

The ISR incidentDiagnosed a revenue drop caused by a rendering change, and learned to treat rendering as a revenue variable.
Rendering has a billMoved from Vercel ISR to a locally-prebuilt pipeline to control cost, hit and fixed OOM at build scale, and settled on SSG / SSR / ISR tuned per site.
Automation with a human gateA tmux auto-loop OOM-ed the machine; redesigned so agents queue work but I approve what runs. Full automation is easy; knowing what to keep manual is the point.
The migrationMoved every asset onto owned infrastructure with canonicals and 301s, absorbed an indexing incident, and recovered rankings.
A sentiment model, 2020A first-year logistic-regression model that classified every review by sentiment and category. It worked, and never shipped.

How I work

  • Understand before abstracting.
  • Measure before trusting.
  • Own what must last.
  • Move the bottleneck, don't fight it.
  • Learn top-down, when the need is real.

LinkedIn to follow. Reach me at wizet1923@gmail.com.