Korea-local products built on LLMs and geospatial data — wrangling Korean map APIs so you don't have to.
Building location-aware products for Korea means building against the Korean mapping stack — Kakao and Naver — rather than the tools most developers reach for by default.
That's the work we do here.
KoPilot began as a travel assistant for visitors to South Korea, and later took part in the ANU TechLauncher program. The problem that started it — tourists can't easily get usable local recommendations — turned out to sit on top of a much deeper set of data and API problems. Those problems are now the substance of the project.
Current areas of work:
- Geospatial data pipelines — ingesting and normalizing Korean place data into queryable form
- Korean map API integration — the practical reality of building against Kakao and Naver
- LLM-backed local recommendations — grounding language models in real place data rather than training-set recall
We publish notes on this work at kopilotapp.com. The early posts are the record of the TechLauncher build.
