Case Studies — Eugine Bhebhe

01 / CASE STUDY

Metups Zimbabwe

A mobile-first second-hand marketplace that brings search, structure, and trust signals to Zimbabwe’s fragmented informal resale economy.

RoleFounder · Solo developer

StackHTML, CSS, JavaScript, Supabase

StatusLive at metups.com

The problem

Second-hand selling relied on disconnected WhatsApp groups, flea markets, and informal networks—without search, categories, trust signals, or reliable discovery across locations.

The product

Metups lets sellers list in under two minutes and lets buyers search by category, location, and price. In-app messaging and seller verification help establish trust before contact details are exchanged.

Engineering

I built the PWA without framework overhead for low-bandwidth mobile use: fuzzy search using Levenshtein distance and trigrams, SmartSuggest, offline capability, and PersonaEngine ranking.

Systems thinking

The product includes 15+ relational tables, RLS, custom PostgreSQL RPCs, a four-tier admin system, commercial analytics, verification, featured listings, and payment-ready monetisation flows.

Outcome

Metups is live in production with the operational, legal, financial, and marketing foundations for a Zimbabwean marketplace. The central lesson: in informal commerce, trust is not a feature—it is the product.

Evidence available: PWA and search implementation, Supabase schema/RPCs, custom analytics dashboard, business and legal documentation.

02 / CASE STUDY

MDUMENI AI Agronomist

An AI-powered crop-advisory and market-intelligence application for Zimbabwean smallholder farmers, developed through the University of Zimbabwe’s INTELLI-Farming initiative.

RoleLead developer · Team of 5

StackReact Native, FastAPI, Supabase

StatusDeployed and active

The problem

Smallholder farmers often lack timely access to agronomists, disease diagnosis, planting guidance, and market information. Poor decisions can directly damage yields and livelihoods.

The product

MDUMENI puts crop guidance, disease information, regional calendars, and market data in a farmer’s pocket. Its name means advisor or counsellor in Ndebele.

Engineering

I led the React Native and FastAPI architecture, implementing clean layers, a 60-crop advisory model, offline-first crop information, and a four-tab market experience.

Critical delivery

I diagnosed duplicate routes, mismatched crop IDs, exposed personal defaults, and unlinked duplicate modules; then delivered marketplace CRUD, RLS-backed tables, and a 10-document technical suite.

Outcome

The deployed system serves 60 crop profiles and includes Pfumvudza basin calculations for 13 crops across Zimbabwe’s Regions 4 and 5—combining cultivation and market decisions in one tool.

Evidence available: API architecture, marketplace schema, mobile screen flow, Pfumvudza calculation reference, and technical documentation.

03 / CASE STUDY

Eloqui Android Tutor

A personal, offline-first Android vocabulary tutor built to make deliberate English learning more useful than passive or overly gamified alternatives.

RoleSolo developer

StackKotlin, Compose, Room, MVVM

StatusBuilt, signed, personal use

The problem

I wanted words in context, adaptive review, useful AI explanations, timely notifications, offline use, and freedom to choose AI providers—without a subscription-driven learning experience.

The product

Eloqui combines a daily word queue, contextual explanations, spaced-repetition quizzes, learning statistics, and configurable notifications in a native Android experience.

Engineering

I built 10+ Jetpack Compose screens on an MVVM and clean-architecture foundation. Room is the primary store, making the app fully usable without a connection.

Design for change

The AIProvider interface gives Gemini, Claude, and Grok equivalent roles behind one contract. WorkManager schedules review prompts around realistic availability windows.

Outcome

Eloqui is a signed app I use personally. It demonstrates that I apply production-minded architecture—replaceable dependencies, local data integrity, and background-work constraints—even when no client requires it.

Evidence available: MVVM/Room architecture, AIProvider interface, Compose screen hierarchy, and notification scheduling logic.