Hojun Park Get in touch

Hojun (June) Park

Search, discovery, and the data behind them.

I own the products behind how customers find what they want: search, recommendations and the product data underneath. In e-commerce since 2016, and in product since 2020.

View case studies Get in touch
Now shippingNew search and recommendations at Carhartt WIP

Where I've worked

Carhartt WIPMytheresaGfK
  • 300KSKUs in the largest catalog I've worked on
  • 25markets at Carhartt WIP
  • 3.5M+searches a month on the platform I'm replacing
  • 6M+customer accounts migrated

Selected work

Six cases from problem to launch, each with the decision that shaped it.

Carhartt WIP, 2026

Replatforming search, browse and recommendations

Own the new discovery platform end to end, from requirements and vendor choice to the MVP and the scale-up across markets.

MVP live in October, full integration by year end

Read case study

Carhartt WIP, 2026

Automating the product data flow

Audited the flow from PLM to storefront, then built automation for seven manual steps with engineering.

Built to remove ~1.5K products a season of manual review

Read case study
Filters on a category page
Size 18,420
Color 17,960
Fit 12,310
Designer 18,105
Occasion 142
Pattern 0
Material 15,870
Sleeve type 38
Neckline 0

Mytheresa, 2024

Better filters from messy product data

Chose to fix the data behind filters, category by category, instead of redesigning them.

+2.3M filter interactions in 30 days

Read case study

Mytheresa, 2023 to 2024 0 to 1

Privé: an invitation-only VIP shop

Took a gated US pilot to a global rollout, letting pilot data decide each next step.

Conversion from 0.2% to 1.2%, six times higher

Read case study

Mytheresa, 2023

Shop migration: 6M+ customer accounts

Owned six months of production readiness across about 20 teams for a new shop system.

25 launch blockers cleared in 3 weeks, pilot on schedule

Read case study

Mytheresa, 2023 to 2024 0 to 1

Fine jewelry: a new high-value product line

Turned dozens of country-specific scenarios into one rule set by value tier.

Live in 3 months, in time for Christmas

Read case study

More work

  • Search personalization pilot

    Mytheresa

    Segmented shoppers by behavior and adapted ranking for new, loyal and premium customers in Germany and the US.

    Click-through +20%, conversion +12% in target segments

  • International checkout

    Mytheresa

    Reworked address and tax handling for new markets.

    Arabic-channel usage from 20% to 35%, Canada live after a 3-day tax fix

  • Incident response

    Mytheresa

    One intake, owner and escalation path for e-commerce incidents.

    Resolution time down 30%, escalations down 40%

Things I've built

Tools I built with AI to take manual work out of my team's week. The Category Builder is now used daily by the team.

Carhartt WIP, 2026

Category Builder

Problem. Seasonal categorization ran on an Excel macro that guessed product types from names. Plain tees ended up under Graphic Prints, and every run needed manual clean-up.

Built. An AI-assisted tool that derives every category from PIM, checks each one against the live category tree, and only removes a category when one attribute proves it wrong. Sale campaigns land in both the campaign and the regular sale tree automatically. Weekly hand-picked categorization is now automatic, and I rolled it out with AI tool training for the team.

In daily use by three team members

Campaign Product Picker

Problem. Campaign edits were picked by hand in spreadsheets, and dates broke on import.

Built. A workflow that proposes the edit, rules out sale and recently featured items, and produces import files with dates intact.

Shared with the team as a versioned skill

Requirement
fabric from description
Akeneo rule
Requirement
color from image
Akeneo rule

AI rule generator

Problem. Writing enrichment rules by hand was slow, and color instructions conflicted.

Built. A generator that turns attribute requirements into Akeneo rules, so enrichment runs on its own. I also rebuilt the product-name rules after tracing a sleeve-length error.

In use: color and fabric enrichment automated

~3,000
styles a year kept current

Outfit links

Problem. Styled-with recommendations were entered by hand and went stale as products changed.

Built. Feeds outfit links straight from photo studio data into the PIM and keeps them current. It also showed the real bottleneck was input quality.

~3,000 styles a year kept current

Product gates
Live
Coming soon, variants
Held back, with reason

Release gates and checker

Problem. Unready products went live, and nobody could see why a product was or wasn't online.

Built. Gates that keep unready products offline, and a checker that shows the reason: product gates, Coming Soon and variant gates.

Errors that used to pass silently are now visible

Stack

The tools I use day to day.

Commerce and data

Commercetools, Akeneo, Constructor, Sanity, Magento

Analytics

GA4, BigQuery, Looker, Power BI, MySQL at Mytheresa

Delivery

Jira, Confluence, Figma

Building

Claude, ChatGPT, Python, JavaScript, Git

Working on search, discovery or product data? Let's talk.

Email me LinkedIn