Currently · Solutions Architect at ColorGen · Founder of Archeta
I build AI systems for real business problems.
I'm a software engineer in Durban. By day I build business applications in C#, ASP.NET Core, React, SQL and Azure; the rest of the time I work on AI engineering, backend architecture, and systems that have to stay reliable once they're in front of real users. I founded Archeta, an AI coordinator that keeps software teams aligned across GitHub and Slack, and I maintain RagCap, an open-source toolkit for portable retrieval-augmented AI. I've entered HackerRank's Orchestrate hackathon three times and climbed every round: 70th, then 5th, then 2nd of 1,983 submissions, one point behind first. I think building something is only half the job. Being able to explain and defend the decisions behind it matters just as much.
Selected work
05 case studiesNotification Router
A personalised message router for WhatsApp. For every incoming message, text, image poster or voice note, it decides whether to interrupt you now, hold it for later, or mute it. The catch is that two people can receive the identical message and the right answer is different for each of them. One agent judges risk, relevance and urgency; deterministic code computes the personalisation and enforces the safety rules.
Evidence Review
A claim-verification agent that decides whether the photos a user submits actually back up their damage claim, across cars, laptops and packages. One reasoning agent inspects each image, weighs it against the claim and the person's history, and returns a verdict: supported, contradicted, or not enough to tell.
Orchestrate Triage
A support-triage pipeline that classifies and routes requests across many companies without ever inventing a policy. It runs on lexical retrieval and rule-based routing instead of letting a model decide freely, so the same request always lands in the same place.
Archeta
An AI coordinator for small software teams. It runs async standups and reads GitHub and Slack into one shared picture of where a project actually stands, flags risk early, and suggests the next thing worth doing. It stays out of prioritisation and out of the code on purpose.
RagCap
Retrieval-augmented generation packed into a single file. You
bundle your sources, chunks and embeddings into one
.ragcap SQLite capsule, then build, search, ask,
serve or export it. It runs offline, gives the same answers
every time, and works from the command line.
How I work
a few things I hold toRules before generation
When a system touches policy, money or production data, I start with rules and retrieval, and only bring in a model where it genuinely earns its place. Predictability is worth more than cleverness there.
Knowing what not to build
A tool is defined as much by what it refuses to do as by what it does. Archeta won't reprioritise your roadmap or write your code, and that line is enforced in the product rather than just promised in the pitch.
Fail loudly, not quietly
I would rather a service refuse to start than corrupt its state in the background. Schema checks on boot, closed-loop verification, no silent migrations. A loud failure is one you can actually fix.
Keep the reasoning
Anyone can stand up a demo. I try to keep the tradeoffs and the measurements alongside the result, so I can explain why something is built the way it is. That is what these case studies are for.
Building software is only half the battle. Being able to articulate and defend your technical decisions matters just as much. — something three hackathons taught me
Contact
open to AI / backend workBuilding something that has to actually work? Let's talk →