Korean property listings are scattered across portals, written in free-text agent notes, and duplicated everywhere. JipData collects them, merges the duplicates, and extracts structured fields — then Claude writes the descriptions agents actually publish.
The same apartment appears on multiple portals in different formats. After collection and deduplication, it becomes a single structured record.
{
"listing_id": "jd_8f3a21",
"type": "apt_jeonse",
"address": { "dong": "잠실동", "complex": "○○자이", "floor": 14 },
"price_krw": 580000000,
"price_per_pyeong": 21400000,
"maintenance_fee": { "amount": 230000, "includes": ["heating", "water"] },
"station": { "name": "잠실새내역", "walk_min": 6 },
"sources": 3, // merged from 3 portals
"agent_note_parsed": true
}
Three pieces of the same system — each usable on its own.
Scheduled collectors pull listings from Korea's major portals. Address, floor plan, price and photo similarity merge the same property across sources into one record.
Claude-powered listing copy in natural Korean. A structured record in — an agent-ready description out, in the office's own tone.
Clean, queryable real-estate data for proptech builders. Search by district, price band, size or station distance — delivered as JSON.
We're a small team in Seoul. This is what we're doing — in order.
Running the collection + dedup pipeline against Seoul districts, expanding coverage area by area. Measuring dedup accuracy before we open anything.
Listing-description generation tested with a small group of agent offices — their listings, their tone, their feedback.
The normalized dataset opened to outside builders, with a field dictionary and honest coverage notes.
Public listings on Korea's major property portals. We collect them on a schedule, merge duplicates across sources, and keep a record of which portals carried each listing.
Agents post the same property on every portal — often with slightly different wording. Without deduplication, any search or analysis counts the same apartment several times.
Korean agent notes are unstructured free text — "관리비 23만 수도·난방 포함, 입주 협의". Claude extracts structured fields from them and generates natural listing descriptions. Rule-based parsing can't handle the variety.
Not yet — we're onboarding early-access partners in small batches while coverage expands. Join the list below and we'll reach out when your region is covered.
Coverage updates, pilot openings and API invites — in your inbox.
Tell us your region and use case — we onboard in batches as coverage lands.
Agent office pilot, custom extraction fields, or data partnership — tell us briefly what you need.
Email us