From the studio. This piece is about the business side of the work, which lives at Mirembe Muse. It is kept here so its link keeps working.
# VarsityOS: Building AI for South Africa's 50% Student Dropout Rate
South Africa has one of the highest university dropout rates in the world. More than half of students who begin a degree don't finish it. This is not primarily a question of academic capability. It is a question of infrastructure.
The student who drops out in month four was not academically unsuited for their degree. They were NSFAS-funded and the payment was three months late. They lived 45 minutes from campus on a taxi that doesn't run after 9pm. They were studying in a flat with no data and scheduled load shedding between 6 and 10pm. They were experiencing the kind of mental health crisis that needed SADAG at 2am, and the campus wellness centre opened at 9am Monday to Friday.
The tools built to support students were built in California. VarsityOS was built for this student.
Who Nova Is
Nova is VarsityOS's AI companion. She is not a study app. She understands NSFAS. She knows the Eastern Cape taxi routes. She can generate a R50 meal from whatever is in your cupboard. She notices when your language has shifted from stressed to in-crisis, and she routes to Lifeline SA.
Nova runs on Claude Sonnet. Her system prompt contains the most comprehensive context I've ever written for a production AI: SA academic calendar, all 11 official languages, NSFAS payment cycle, university-specific terminology, cultural references that make her feel like she belongs in this context rather than imported into it.
The Six Agents
VarsityOS deploys 6 independent AI agents on a shared Supabase database:
- Nova (Core Companion): Conversational AI with full student context. The daily interaction point.
- Study Planner Agent: Generates adaptive timetables, tracks module progress, alerts to upcoming deadlines.
- Budget Agent: NSFAS tracker, expense categorization, weekly budget summaries with realistic alerts.
- Meal Agent: Generates recipes from available ingredients for a specified budget. Knows what R50 can actually buy at a township spaza shop.
- Crisis Detection Agent: Monitors language for mental health signals. Routes to SADAG (0800 567 567) or Lifeline SA (0861 322 322) when the signal threshold is crossed.
- Wellness Check-in Agent: Daily morning check-in. Tracks stress levels, sleep, food. Escalates to management contact if 3+ consecutive crisis-level responses.
Six agents. One database. Each agent knows nothing about the others' conversations, but can access the shared state (module list, budget baseline, crisis history) that persists across them.
The PWA Decision
VarsityOS is a Progressive Web App. Installable. Offline-capable. This was not an accident.
A student with 200MB of data left doesn't open a new browser tab. They open an app icon. VarsityOS installs to the home screen on first visit with a single tap. Every cached page is available offline. Critical information (study schedule, budget summary, emergency numbers) loads from local cache before network.
On first install, we pre-cache the crisis contacts. If a student is offline and distressed, they still get the Lifeline SA number. This is not a product feature. It is a moral requirement.
Paying Users and Cultural Validation
200+ active students in beta. This is not a large number. But it is meaningful: these students chose to install, to create accounts, to invite Nova into their daily routine. In a market where students have been failed by institutional tools, voluntary adoption is the highest validation.
The Free tier intentionally includes crisis detection. You do not need a subscription to have the crisis agent active. This is a decision I made deliberately: no student should be blocked from the mental health layer by a payment that hasn't cleared from NSFAS.
What We Learned
Context specificity is everything. A generic study app gives generic advice. Nova gives advice that accounts for that this student's NSFAS allowance arrives on the 15th, that they have an exam on the 18th, and that they mentioned not eating yesterday.
Offline-first is care. Every feature decision was run through the filter: what happens when there's no internet? The answer shaped a fundamentally different architecture than a standard web app.
Crisis infrastructure is not optional. If you're building mental health adjacent tools for SA students and you don't have crisis routing, you are not building a product. You are building a liability.
VarsityOS is proof that the technical skills to address South Africa's deepest social problems are here, on the continent, in the hands of developers who understand the problems from the inside.
What would South African higher education look like with AI designed for it, not borrowed for it?
Reader Insights
0 responses
No insights yet. Be the first!