Technical notes on data engineering, machine learning and governed AI, from Harare.https://petermundowa.com/en-gbhttps://petermundowa.com/blog/measures-and-calculated-columns/https://petermundowa.com/blog/measures-and-calculated-columns/The most common Power BI mistake I see, why it produces confidently wrong numbers, and the one question that tells you which to use.Sat, 18 Jul 2026 00:00:00 GMThttps://petermundowa.com/blog/profile-the-data-before-you-trust-it/https://petermundowa.com/blog/profile-the-data-before-you-trust-it/Five minutes of profiling saves five weeks of explaining. The checks I run on every new table before a single chart gets built.Sat, 18 Jul 2026 00:00:00 GMThttps://petermundowa.com/blog/accuracy-is-a-lab-measurement/https://petermundowa.com/blog/accuracy-is-a-lab-measurement/A model that scores 95 percent on a clean test set can fall apart on a cheap phone in bad light. Robustness is a training decision, not an afterthought.Mon, 06 Jul 2026 00:00:00 GMThttps://petermundowa.com/blog/an-empty-field-is-never-empty/https://petermundowa.com/blog/an-empty-field-is-never-empty/NULL, zero, N/A and a blank string look identical on screen. They are four different stories, and your pipeline needs to know which one it is reading.Mon, 06 Jul 2026 00:00:00 GMThttps://petermundowa.com/blog/keep-the-model-away-from-the-numbers/https://petermundowa.com/blog/keep-the-model-away-from-the-numbers/The reliable architecture for AI systems on business data is simple: the database retrieves, code calculates, and the model only writes the sentences.Mon, 06 Jul 2026 00:00:00 GMThttps://petermundowa.com/blog/the-model-has-never-been-the-hard-part/https://petermundowa.com/blog/the-model-has-never-been-the-hard-part/What enterprise AI actually looks like when the schema is undocumented and the person who understood it has moved on.Wed, 01 Jul 2026 00:00:00 GMT