Projects

The SA-CDI project is gathering Communicative Development Inventory responses from caregivers across South Africa to establish early-language development norms in the country's official languages, with a target of roughly 20,000 responses. An earlier pilot used a Flutter app, but asking caregivers to install software created real barriers around data cost, device storage, and digital literacy. My project delivers the same questionnaire through WhatsApp, a platform most caregivers already use, with no installation at all.

I'm building the system in Python and Flask, using Twilio for WhatsApp messaging, Firebase for storage, and soon Azure neural text-to-speech wil be used so caregivers with lower literacy can hear each question read aloud. It runs in English, Afrikaans, and isiXhosa, persists session state server-side so a questionnaire can be paused and resumed across conversations, and pseudonymises participant data to stay POPIA-compliant. Alongside the engineering, the work is structured around three research questions: how the WhatsApp interface compares to the original app, which text-to-speech system serves each language best, and how the interface holds up under usability evaluation.

We developed a WhatsApp chatbot with a monitoring-and-evaluation dashboard for an NGO supporting caregivers in their child's first thousand days. The chatbot delivers approved parenting and health content in English and isiXhosa through a menu-based, on-demand model, so it never sends unsolicited messages.

It answers frequently asked questions using a retrieval-augmented generation pipeline constrained to vetted sources, with content curated from other NGOs (Wordworks and Side-by-Side), and flags out-of-scope or clinical questions for human review rather than answering them itself. The backend runs on Flask and SQLite with Twilio for WhatsApp integration, and the React dashboard gives staff analytics on engagement, question categories, and retention to support evidence-based decisions. We worked in a user-centred, co-design process with NGO (Layita) staff, with attention to anonymisation, consent, and cultural sensitivity.

In my final year of undergrad, we built the Campus Carpool App for Gendac, a software company that brought us the brief as a real client. They wanted an Android app that connects Stellenbosch students with similar travel routes and schedules so they can share lifts - splitting fuel costs, cutting emissions, and easing the pressure on campus parking, while giving students a safer and more flexible alternative to driving alone or relying on public transport.

We built it with .NET MAUI and C# on an Azure backend, using Google Maps for routing and distance estimation and GitLab for version control and issue tracking. Students verify with their university email and an OTP, then set themselves up as a driver or passenger, lay out a route with departure and arrival windows, and get matched with compatible students. The driver then adds the cost so that the potential passanger can view the cost of the ride, matched users are connected through WhatsApp.

I built this site because a CV and a GitHub profile only tell part of the story. I wanted somewhere I could show the projects I've enjoyed working on, explain the thinking behind them properly, and let a bit of my personality come through; not just a list of technologies I've touched.

It's built with Astro and Tailwind CSS v4. I designed it to start from a clean, simple foundation and grow more expressive over time, layering in details like the polaroid-style photos and fade-up transitions once the basics were solid rather than trying to get everything perfect in one pass.

As a scientific-computing intern at SANSA, I worked on a Python codebase implementing Spherical Elementary Current Systems (SECS). SECS is a technique for modelling ionospheric currents from the geomagnetic field measured at ground observatories, in this case across South Africa and Namibia. My focus was making the existing scientific code cleaner, more configurable, and faster to run.

I refactored what had been a single script into modular, configurable functions, using dataclasses and config files to separate the science from the parameters so runs could be set up without editing the code. I implemented the conversion of raw magnetic field components into the form the SECS calculations required, and optimised the data-processing pipeline by moving it onto vectorised pandas operations rather than slower row-by-row handling.

Lastly I translated the data onto a visible map, using Cartopy. This allowed one to view the direction and magnitude of the magnetic currents on a map of South Africa.

CSV-Flow is a Java-based desktop application for building CSV data workflows - a tool aimed at end-user productivity, with support for configurable actions, scheduling, and licensing and updates. I joined as a freelance developer for six months to help maintain and extend it, working remotely for 29CPmax.

I started by getting to grips with an existing codebase, tracking down and fixing bugs to improve stability and performance before moving on to new functionality. From there I built out features that extended what the tool could do, including improvements to the workflow editor and to the user-facing operations people rely on day to day. The application is built and packaged with Maven.

BE RUG BE is a French rugby ball brand, and I'm designing and building their showcase website as a favour for a friend. Rather than a full storefront, the site focuses on presenting the range clearly and sending visitors out to purchase, so the emphasis is on clean product presentation and brand feel rather than checkout logic.

It centres on the three flagship balls - the Alpha-T5 for training, the Elipse for match play, and the Shark for competition - each with its own presentation and an outbound link to buy. It's built with Next.js using the App Router and TypeScript, styled with Tailwind CSS, deployed on Vercel, and version-controlled through GitHub. The foundational architecture and component structure are in place, with content and asset integration still ongoing.