Putting the right worker on the map, in minutes

Each list query is part radius/bounding-box filter (who is near this point) and part marketplace state (job status, application history, proposals). The relational joins — job to business to address to application status — dominate the workload, and we needed real Earth-distance math, not flat-plane latitude/longitude.

So we stored each address as a PostGIS geography point on Postgres and used ST_DWithin for the "near me" radius and ST_MakeEnvelope / ST_Within for the map's pan bounding box — a single query powers both the nearby list and the live map view.

We chose Kysely over Prisma because the geo predicates only express well in raw SQL, and Kysely lets you drop into a typed SQL block without losing type safety on the rest of the query. We passed over MongoDB's $near and managed search services like Algolia or Elastic, because the relational joins dominate and we wanted true geography, not approximation. Every list is cursor-based, not offset, because new jobs and applications are inserted constantly — and offset pagination silently skips and duplicates rows while a user scrolls.