The problem
Many places have no reliable address system. A friend's street in Laos has no address, so there is nothing consistent for a map, a delivery or a form to point at. Project:Address generates those addresses from data that already exists.
An AI-powered addressing system for places without reliable addresses. It generates block and building numbers from OpenStreetMap and official boundaries, starting with Laos.
Jan 2025
started, while I was serving in the army
Laos
the first region, where a friend's street has no address
4 levels
of official boundaries it works within: country, province, city, village
30+
founders and engineers I showed it to at SF and LA Tech Week
Many places have no reliable address system. A friend's street in Laos has no address, so there is nothing consistent for a map, a delivery or a form to point at. Project:Address generates those addresses from data that already exists.
Before designing the numbering I read how addressing is already defined elsewhere: the ISO 19160-1 address model, the UPU S42 postal standard, and national profiles such as the UK's BS 7666, the US Postal Service's Publication 28 and New Zealand's address profile.
Everything happens inside an administrative unit, loaded from official boundaries one level at a time: country, then state or province, city and village. OpenStreetMap supplies the roads and buildings within it.
Blocks are generated automatically from the road network, with density heuristics. Buildings are then numbered grid-first, with adjustments that follow the roads.
The address strings are deterministic and built on conflict-free rules, so the same building always gets the same address.
Rules first, learning-based over time. The numbering starts from explicit rules that can be checked, and the plan is to improve how addresses are ordered and generated with feedback from people who use them.
projectaddress.dev runs a web app with search and a country view over sample buildings. The project is on hold during my studies, and I plan to resume it.
Next: stabilise the hybrid numbering so no address repeats, improve the AI ordering with user feedback, and open an API and a Data Studio.
Frontend
Next.js, React, Tailwind CSS and Leaflet.
Backend
FastAPI (Python).
Data & GIS
PostgreSQL + PostGIS, OpenStreetMap and official boundaries.
Want a deeper dive or a demo?