Most of the difficulty was never the analysis.

I trained as a surveying engineer at the University of Zimbabwe, which is why I think in error budgets, ground truth and validation samples rather than in demos. Almost every hard problem — land rights, lending, flood response, infrastructure planning — has a geography to it, and most of the difficulty is not the analysis. It is reconciling identifiers and trusting the data underneath.

That became a career architecting the systems Zimbabwe's institutions now run on. I was the GIS systems architect on the team that delivered the national electronic cadastre, and on the team behind the first GIS-driven mortgage platform in Zimbabwean banking. Both are live and in daily use. I did not build either alone, and neither is mine — but the spatial architecture on both was my work.

My own projects are where I get to be exact about what is and is not true. A QGIS plugin that packs a multi-gigabyte orthomosaic into one shareable file. A crop-disease pipeline where the vision model is forbidden from naming a disease and a build-time validator fails the build if an output recommends a banned pesticide. A building-change record for one Harare suburb that reports, in its own README, the one thing it cannot detect.

Where I am going

The slow part of this work was never the analysis — it was hunting down data and reconciling identifiers that nobody designed to match. That is the part I now hand to machines, carefully and with the result measured. Retrieval that grounds a model in verified sources instead of its own priors. Evaluation harnesses that report a number rather than an impression. Validators that fail a build when an output is unsafe.