Issue 003 · A Career Feature · Byline — Herself
The Making
of an Engineer
From a single line of code to a fifteen-wing AI operating system — in one year. The other half of the poet who codes, told the way it deserves to be told.
Contents
In This Issue
FIVE CHAPTERS · ONE RISE
I.Chapter I · July 2025
Zero.
There was a version of this story where it never started. No computer-science degree. No bootcamp cohort. No mentor down the hall. Just a business graduate in KuGompo City who opened an editor and typed the first line she'd ever written.
She didn't come to code from theory. She came to it from need — the products in her head had nowhere to live until she could build them herself.
“I learned to speak in two tongues — the language of systems, and the language of longing.”
— N.R.K-K.
Chapter II · The Sprint
Eight apps.
One year.
Not eight ideas. Eight shipped, production, paying-user applications — each a rung on a ladder that got steeper on purpose. Read it top to bottom: this is a year measured in deployments.
First line of code
HTML · CSS · JavaScript
SheCodes Plus · first deployed builds
First production app shipped
Next.js · TypeScript · Supabase · PayFast
Cortex Hub · GreenVault
Geospatial + real users
Mapbox · PostGIS · Expo · complex data models
True Access v1 · PoetryTube
AI engineering begins
Claude Sonnet + Haiku · prompt caching · multi-agent
VarsityOS (Nova AI) · StokvelOS
Enterprise-grade architecture
Inngest · Upstash Redis · WhatsApp API · RLS at scale
AdminOS · K53 Drill Master · WatchSankofa
15-wing AI OS + mobile co-founder
Expo SDK 52 · RAG · Redis inter-wing signals · iOS/Android
JarvisOS · Sanyu Botanicals · True Access v2
III.Chapter III · What she builds now
The Craft.
The distance between the first chapter and this one is the whole point. She no longer builds pages — she builds systems that think: agents that route work between themselves, retrieval layers that answer from thousands of chunks, cost architectures that make production AI actually affordable in rands.
This is what “I've evolved” looks like on paper.
Multi-Agent Systems
Specialist agents with clear mandates, an Inngest async spine, and a Redis signal protocol so fifteen wings talk without stepping on each other.
RAG + Vector Retrieval
Knowledge ingestion pipelines and document Q&A over Upstash Vector — 1,194 live chunks answering in production inside JarvisOS.
Model Routing + Caching
Sonnet for reasoning, Haiku for speed, prompt caching on every static block. An 85% cost reduction, engineered — not hoped for.
Multi-Tenant Architecture
Supabase Postgres with Row-Level Security, soft-delete and immutable audit trails. One deployment, unlimited tenants, isolated by privilege.
WhatsApp-Native AI
Meta WhatsApp Cloud API workflows — the channel Africa actually runs on — automating inbox, debt recovery and reporting inside AdminOS.
Production TypeScript
Strict mode, Zod at the edges, tsc --noEmit exits zero across every repo. The discipline is the point.
The Numbers
Receipts, set in gold.
8
Live production apps
3,000+
GitHub commits
1,194
RAG knowledge chunks
15
JarvisOS intelligence wings
15
Academic distinctions
3+
Paying clients
Chapter IV · The Proof
Not a portfolio.
A production record.
Every project below is live, self-funded, and carrying real users or real revenue. No demos. No mock data. The four here are the ones that best show the range — from a fifteen-wing personal OS to a universal mobile platform shipped in two days.
Chapter V · The Foundation
The business brain
under the code.
Three qualifications from Nelson Mandela University. Fifteen distinctions. People assume the engineering came first. It didn't — the strategy did. Every architectural decision has a business module in its DNA.
Advanced Strategy
→ Eight-product horizontal architecture — one payment hub, six apps downstream.
Operations Management
→ AdminOS immutable audit log — UPDATE and DELETE revoked at the database privilege level.
Business Ethics
→ POPIA compliance from day one. Soft delete only. Timestamps always UTC.
Financial Literacy
→ Per-tenant ZAR token budgets with 50% hourly spike detection — AI cost control in the data model.
The Back Cover
This is what I build
for clients.
Custom Claude agents. WhatsApp-native workflows. Production TypeScript. Built for the continent, available to the world.