Cookie

Cookie consent

We use some essential cookies to make this website work. We'd like to set additional cookies to help us measure your experience when you view and interact with the website.

Cookie policy
Pricing Log in Start a free trial

Using AI to see our address data differently

One developer and two hours with AI produced an interactive 3D view of millions of addresses to help us check the data behind an upcoming Postcoder improvement.

Using AI to see our address data differently
Daniel Cooper Published 4 Nov 2026 Product

We're working on an enhancement to Postcoder address autocomplete: optional proximity biasing. When enabled, this will help people find their address by taking their current position into account alongside what they're typing.

Behind the change is some interesting work with address data and a new internal visualisation tool that helps us see what tables and test results alone can't easily show.

Exploring the address data

Our visualisation tool lets us explore address datasets as interactive 3D trees. The screenshots here show UK data, with branches connecting counties to localities, streets or postcodes and individual premises.

The Isles of Scilly, with the Cornish coast in the background. Viewed from the edge, the full, unfiltered tree reveals its layers: counties, localities, streets and postcodes, then premises.
The Isles of Scilly, with the Cornish coast in the background. Viewed from the edge, the full, unfiltered tree reveals its layers: counties, localities, streets and postcodes, then premises.

County Fermanagh in Northern Ireland shows how varied address structures can be. Many rural addresses are organised around named areas called townlands rather than streets, giving the tree a different shape from a typical town or city.

County Fermanagh, showing its distinctive townland structure within the address hierarchy.
County Fermanagh, showing its distinctive townland structure within the address hierarchy.

Zooming into Norfolk shows how individual addresses connect to streets and the villages they belong to, making it easier to follow each branch and check that the relationships make sense.

A closer look at part of Norfolk, showing the address tree branching from village localities down through streets to individual premises.
A closer look at part of Norfolk, showing the address tree branching from village localities down through streets to individual premises.

Moving between the national picture and individual branches helps us spot anomalies and choose test cases, complementing our automated checks.

County Fermanagh again, giving a sense of the number of nodes involved.
County Fermanagh again, giving a sense of the number of nodes involved.

Two hours rather than several days

Kelvyn Scrupps, one of our senior developers in DevOps, built the tool in around two hours using Cursor, an AI-powered code editor. He used Python and JavaScript, with Three.js for the 3D visualisation.

Across the team, AI is making tools like this practical to build in hours rather than days. That makes it easier to try ideas that might previously have taken too much time to justify, even if they're only useful for one project.

This one is helping us check the data behind proximity biasing. It also gave us a view of our address data we thought was worth sharing.