An open, version-controlled dataset of AI-relevant datacentre infrastructure across Africa — 14 facilities in 5 countries, every data point sourced, dated, and assigned a confidence level.
Some of the work that I have done, either to scratch an itch, to learn, or for clients.
In the field
Systems deployed for real people and real institutions.
Standards & open source
Changes landed in codebases and specifications I did not start.
Editorial and code contributions to the WHATWG HTML Standard, the Matrix protocol website, and Inspect — plus a TC39 proposal implemented in SpiderMonkey.
Reading infrastructure · 2021–present
One corpus, five years: 14,257 highlights and notes from 299 books, 2019–2025.
A Node.js CLI that parses a Kindle's clippings file into a folder of Markdown.
Zhira, Shona for “path(s)”, is a versioned context pipeline over a reading corpus, turning Kindle highlights and notes into a retrievable, reliable, cited context for humans and models. Early build.
Extraction → organisation → retrieval. Each layer only became worth building once the one beneath it was solid.
Tools & experiments
Built to learn something, or because I wanted the tool to exist.
An Electron desktop app that imports, parses, and organizes your Kindle highlights so they stay useful after you finish a book.
A neural network implemented from scratch with vanilla JavaScript and trained on the MNIST dataset.
Client work
The trade that funds the rest.
Website for the Women's Academy for Leadership and Political Excellence Africa, a continental organisation training and mentoring women for political leadership.
Built and maintained public-facing web infrastructure as part of a broader Tech Fellow relationship.
A website development project, using ReactJS and the Astro framework.
Practical tutorials and guides for TalkJS's chat SDK, published on their developer resources site.