A land valuation tool

The situation

A client wanted an automated valuation tool for urban leasehold plots in Addis Ababa, for landowners and brokers. It estimates a price per square metre from government lease-auction records and the ground around each plot, and it explains every figure it gives.

What I did

On the first day I had the brief and a six-row sample of the data, and four throwaway UI prototypes were built from them: a calculator, a guided wizard, a map-first atlas and a printable report. The atlas was chosen the same day. You drop a pin on the map and the valuation and comparable sales appear around it. There were 17 logo concepts too, and the one chosen is based on the Adey Abeba, the flower Addis Ababa is named after.

On the second day the spec was written around the atlas, and then the build: the valuation engine, the API, the React web app with all six screens, an end-to-end walk of the deployed site, and a security pass that fixed what it found. The interface is in English, Amharic and Afaan Oromoo.

The model is ridge regression on the 2020 to 2026 auction records, with features from OpenStreetMap for the ground around each plot. No GPU, and no per-query AI bill. For a plot it gives a price per square metre, a likely range, the comparable sales it used, and how much each factor moved the figure.

The result

A working tool in three languages, with 246 automated tests. It shows an "indicative, demo data" label, which was to stay until the full dataset and a valuer's sign-off were in place. The project has ended and the code is archived.

What a team can take from it

Now we can talk to customers through quick prototypes much more easily. People who don't code don't really understand a spec, and a working prototype lets us communicate with customers in a much more real way. Of course this means more code gets thrown away, but I think the final result will be more meaningful and closer to what the customer has in mind.

Faster iterations, more customer satisfaction and a better time to market should be the result. When we build with a good methodology and quality testing, we can be fast too. I guess it's "build fast and break fewer things" now.