Work & Projects

Every data team rebuilds the same pipelines, and the work dies whenever the platform changes. And too much of the tooling Africa runs on is built elsewhere. Ubunye, isiZulu for unity, exists to produce serious open source from here.

An open source organisation whose flagship, Ubunye Engine, lets you describe a data or ML pipeline once, as a small folder of config and Python, and run that exact folder on a laptop, Docker, Kubernetes, cloud clusters, or Databricks. For data and ML teams who want pipelines that outlive their platform, and for learners who want production habits from day one.

Released on PyPI with the same pipeline proven to produce identical output on seven different environments, eleven worked examples, and a full documentation site.

Real sensors lie. Vehicle trackers, IoT devices and industrial equipment all produce noisy readings, and you need an honest estimate of what is actually happening underneath the noise.

An open source Python library of Bayesian filters, including Kalman, Particle and Ensemble filters, with a familiar scikit-learn style API that scales out with Dask. For engineers and scientists working with noisy time series who want proven mathematics without writing it themselves.

Published on PyPI and used for telematics scoring, IoT forecasting, and real time state estimation at scale.

Two problems with one root. Township businesses stay invisible online because agencies cost more than they can spend. And township youth cannot turn their potential into income, because the skills that now pay, building with AI, are taught everywhere except here.

An initiative that teaches young people to build with AI by doing real work: they use AI to ship real websites and tools for township SMEs, described in plain English, with no formal coding background needed. The businesses get a digital storefront and new customers; the young builders get in-demand AI skills, a public portfolio, and a path to earning from it. Built to create employment and grow the local economy, not to hand out charity.

Seven live sites already shipped for real businesses at no cost to them, including restaurants, a school, a cleaning service, a photographer and a financial advisory, each one built as hands-on training for a young person learning to earn with AI.

Load shedding forced thousands of network sites onto diesel generators, and every bad dispatch decision burned fuel and dropped the network for real customers.

A constrained optimisation engine with real time streaming analytics that decides which generators to run, where, and when. Built for the network operations teams who keep South Africa connected through power cuts.

Reduced downtime by 5%, lowered operational costs by 30%, and supported R1B in annual savings.

Climate and environmental risk questions need planet scale satellite data, and almost no organisation can process it on its own.

Analytics workflows on IBM PAIRS that process multi terabyte satellite raster and vector datasets. For enterprises that need environmental, climate and supply chain risk intelligence they can act on.

Integrated into IBM's Environmental Intelligence Suite for global environmental monitoring.