Work & Projects

ABSA lacked consistent ML/ETL standards, reusable engineering patterns, and scalable workflows.

Built a modular open-source ecosystem providing declarative ETL engines, ML pipelines, feature stores, and YAML-driven orchestration.

Accelerated model delivery, improved data quality, and uplifted engineering capability across teams.

IoT and telematics pipelines required robust filtering for noisy sensor data and real-time state estimation.

Created an open-source Bayesian filtering library supporting Kalman, Particle, and Ensemble filters with distributed execution.

Improved reliability of IoT forecasting, telematics scoring, and real-time analytics at scale.

Township businesses lacked affordable digital presence and branding tools.

Developed scalable web templates and branding systems deployed on cost-free hosting.

Enabled SMEs to reach customers online and strengthen brand identity.

Load-shedding caused fuel inefficiency, network downtime, and poor generator deployment decisions.

Developed a constrained optimisation engine with real-time streaming analytics to prioritise generator dispatch.

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

Enterprises required scalable geospatial intelligence for environmental, climate, and supply-chain risk.

Built analytics workflows using IBM PAIRS to process multi-terabyte raster and vector datasets.

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

Public health teams needed real-time visibility of COVID-19 hotspots.

Built an interactive dashboard with automated heatmaps and case-tracking visualisations.

Used by Gauteng Department of Health for hotspot identification and rapid response.

Municipal managers lacked predictive tools for service delivery optimisation.

Developed ML models (SVM, Random Forest, PCA) and delivered insights via a Django dashboard.

Improved municipal planning and youth unemployment analytics for the City of Cape Town.

First-year students struggled to choose optimal courses, leading to poor completion rates.

Built a K-modes clustering recommendation engine deployed via an interactive dashboard.

Achieved 90% silhouette score and enabled more informed decision-making for students and advisors.

Iterative segmentation using RNNs was computationally expensive.

Designed an Echo State Network architecture enabling efficient segmentation with reduced training overhead.

Published research demonstrating competitive performance at a fraction of compute cost.

South African students lack accessible tools to navigate university applications and job hunting. Most processes are manual, fragmented, and overwhelming.

Built a WhatsApp-based AI agent that reads matric certificates via Claude Vision, calculates APS scores, matches students to eligible university programs, and submits applications automatically. Extended to job applications with CV parsing and automated form submission.

Enables students to apply to universities and jobs by simply messaging a WhatsApp number. No app download, no website, no friction.