.jpg&w=3840&q=75)
How we build
Every system here is a response to a human problem, a market reality, and a business constraint, at the same time. Here is the thinking behind the work, in three lenses.
The business lens
~68% lower LLM cost, the difference between negative margin and a viable B2B product at scale.
The business lens
VarsityOS keeps a duty of care to its users with no dependency on a third party staying online.
The business lens
A cache outage never becomes a user outage, no SLA breaches from infrastructure hiccups.
The business lens
A genuine learning-outcome differentiator with zero per-seat licence cost.
The business lens
Enterprise-grade tenant isolation at no additional cost.
The business lens
Meet the market where it already is, no behaviour change required to adopt.
.jpg&w=3840&q=75)
Build journeys
Every project has a turning point — an architecture call that determined whether the economics worked, whether the product survived real users, or whether the client could own it for life. These are ours.
Personal Operating System
A 27-wing OS that never blocks the user.
Every background job (AI calls, Notion sync, social publishing, email parsing) runs through Inngest, not inline. User-facing requests return immediately; the work happens behind the curtain. 254 workers, zero UI lockouts.
Build something like this
AI Intelligence SuiteWhatsApp Business OS
Enterprise-grade B2B OS, native to WhatsApp.
Prompt caching (cache_control: ephemeral) separates the static system context from per-request user data. The first call caches the system block; every subsequent call reuses it. 68% lower LLM cost. The architecture was priced before it was built.
Build something like this
Web + Chatbot BundleStudent Study Platform
Crisis detection at zero latency. No third-party dependency.
Crisis detection runs local pattern-matching before any AI call, so it fires in under 2ms regardless of whether the AI provider is up. Spaced repetition (SM-2 algorithm) was built in-house — no per-seat licence cost, and actual retention outcomes instead of a third party's black box.
Build something like this
AI Starter KitCustom Publishing Platform
93 pages. Built, not bought.
Off-the-shelf platforms (Substack, Ghost, WordPress) compromise editorial control and long-term ownership. The decision was to build from the ground up: custom content types, a reading experience that matched the magazine's identity, and a publishing workflow the editorial team could own without a developer on call.
Build something like this
Custom Platform Build.jpg&w=3840&q=75)
Build teardowns
Every system above started with a choice. These are the full breakdowns — the context, the alternatives rejected, and why the decision that won was the right one.
The hardest engineering problem wasn't the AI. It was preventing 254 background workers from degrading the user interface. Here's how Inngest changed everything.
The decision
“Queue everything. Return immediately. Sync behind the curtain.”
AdminOS had to work at R2,500/month. That meant the LLM cost had to be a fraction of what Anthropic normally charges. This is the caching architecture that made it possible.
The decision
“Cache the static context. Send only what changes. Reuse the rest.”
When you build for students, you carry a duty of care. But AI providers go down. Here's how we built a crisis detection layer that fires before any model call.
The decision
“Pattern-match first. Model second. Never invert this order.”
Ghost, Substack, WordPress — all fine platforms. None of them let a literary magazine own its identity completely. Here's the case for building from the ground up.
The decision
“Own the stack. Own the identity. Own it for life.”
Live products
Every product below is live, paying, and maintained. Click through and try them. This is what your project could look like six to twelve weeks from now.
WhatsApp Business OS
LiveThe business OS that lives where your clients already are.
Multi-tenant SaaS for African SMEs. WhatsApp-native CRM, AI chat, invoicing, booking and staff management — all in one.
R2,500
entry pricing
68%
lower LLM cost
native channel
Student Study Platform
LiveAn AI study OS for students who cannot afford to fail.
Spaced repetition engine, past paper drill, AI tutor and crisis detection — built for South African students on data-poor connections.
150+
live users
< 2ms
crisis detection
SM-2
spaced repetition
Learner Licence Prep App
LiveThe fastest way to pass K53 in South Africa.
A mobile-first drill app with adaptive question selection, progress tracking and a practice test that mirrors the real exam.
Mobile
first architecture
Adaptive
question engine
K53
compliant
Custom Publishing Platform
LiveA literary magazine that owns every pixel of itself.
Custom editorial CMS, reading experience and audience engine built from scratch — 93 pages, zero platform dependency.
93
pages live
24
articles in 90d
100%
client-owned
.jpg&w=3840&q=75)
Built in public
Every build journey has a commit history behind it. Not a slideshow — a changelog.
GitHub
Repository stats are available on request. Every project ships with a full commit history and architecture decision log.
github.com/Nanda-RegineBuilt under constraint
Load shedding
Offline-first PWA architecture, the product keeps working when the power does not.
Data poverty
Aggressive caching and compressed assets, built for R-priced data bundles.
No credit cards
PayFast over Stripe, built for the majority without international payment cards.
WhatsApp as infrastructure
Not a feature, a thesis. African businesses operate where they already are.
