DEV Community: Agbor Thomas Ayuk Carmelo

Mobile money is the API you're missing: shipping payments in Cameroon with CamPay

Agbor Thomas Ayuk Carmelo — Thu, 08 Oct 2026 09:12:42 +0000

Every tutorial assumes your users have a credit card. In Cameroon, that assumption quietly drops most of your market before the first payment screen ever renders.

I build small AI products for the Cameroonian market, and both of them — CVBoost (AI CV tailoring) and WardWatch (civic issue reporting) — take payments through CamPay, a gateway that connects your app to MTN Mobile Money and Orange Money. No cards required.

The product decision that changes everything

The usual path is: Stripe → "we don't support Cameroon" → a side quest into payment politics. The pragmatic path is CamPay.

CVBoost lets users pay per use with MTN MoMo or Orange Money.
WardWatch takes city subscriptions through the same rails — plus an ads marketplace and promotions, all mobile-money backed.

Same gateway, two very different products, one lesson: in Cameroon, mobile money is the payment API. It has ~tens of millions of users and it's how people actually pay for things every day.

What you touch when you integrate mobile money

Compared to card rails, the pattern you're signing up for is a redirect flow:

User hits "pay" → your backend creates a payment request
They authorize on their phone (USSD or app) with MoMo or Orange Money
A webhook confirms the transaction → you fulfill the order

The parts that quietly take time: sandbox testing, idempotent webhook handling (the same webhook will arrive twice), and receipt UX that matches what users expect from mobile money — short confirmations, not invoices.

Payments are a feature your AI product inherits

Neither product ships payments as an afterthought bolted on after launch:

CVBoost tailors a CV against a job description using Groq's LLaMA 3.1 70B — paying users are waiting mid-application, so inference speed is part of the product.
WardWatch's flow includes AI categorization, priority scoring, duplicate detection, and moderation before anything is ever shown on a map.

The paid tier existed before the product was polished to death — because in this market, the ability to pay is a feature, not an admin chore.

Practical notes if you do the same

Offer the local payment method first — card support is a nice-to-have, mobile money is the default.
Webhooks are messages, not commands — make fulfilment idempotent.
Localize the receipt — French-first copy for Cameroon beats English boilerplate.
Price in the local economy — per-use pricing with mobile money beats monthly subscription friction.

Both products are live if you want to see the flows:

CVBoost: https://cvbuilder-obfc.onrender.com
WardWatch: https://wardwatch-taap.onrender.com

Feedback welcome — especially from developers who have shipped payments in markets their payment-provider docs forgot.

WardWatch: turning photos and voice notes into a live map of what a city actually needs

Agbor Thomas Ayuk Carmelo — Thu, 08 Oct 2026 07:49:30 +0000

Most civic complaints never become data. They sit in a chat thread, get forwarded twice, and vanish — while the same pothole is reported forty times and nobody can say which street is worst.

WardWatch (formerly CommunityPulse) is an AI-powered civic intelligence platform connecting citizens, local government, and service providers. The loop: a citizen reports an issue with GPS coordinates, a photo, or a voice note. AI does the boring part. Everyone else sees a live map.

What the AI actually does

Reports flow through LangChain custom workers calling Gemini and OpenAI — four jobs in one pipeline:

Categorization — what kind of issue is this?
Priority scoring — how urgent is it relative to everything else?
Duplicate detection — forty reports of one pothole collapse into one record
Content moderation — keep reports readable

Deduplication is the part most reporting tools skip. It's what turns a complaint box into a dataset a city can act on.

From reports to a map

Processed issues land on real-time heatmaps with geospatial clustering (Leaflet, MapLibre GL, Deck.gl). City admins get a dashboard with analytics for their area, and access is role-based: citizen / admin / superAdmin / city_admin, so a resident and a city official never see the same controls.

The platform also includes communities, events, real-time chat, and web push notifications — it isn't only a complaint queue.

The unglamorous details

Stack: React 19, Vite 7, Chakra UI up front; Node/Express 4, Mongo 8, Redis, Pinecone, Socket.io, and Bull queues behind it; Docker, Sentry, and Cloudinary for the plumbing.
Payments: CamPay mobile money for city subscriptions, the ads marketplace, and promotions — cards are not the default in the market this is built for.
Audience: Cameroon-first: citizens, local government, service providers.

Takeaway

A tool that only collects complaints creates a backlog. One that categorizes, scores, deduplicates, and maps them creates a work queue.

Try it live: https://wardwatch-taap.onrender.com

Feedback welcome — especially from anyone who has tried to get a pothole fixed.

CVBoost: tailoring a CV to the actual job description with Groq's LLaMA 3.1 70B

Agbor Thomas Ayuk Carmelo — Thu, 08 Oct 2026 07:49:28 +0000

One CV, fifty applications: it's the default strategy, and it's a losing one. Recruiters read for fit against a specific job description; a generic CV reads as generic. Tailoring by hand works, but an hour per application adds up fast when you're applying to ten jobs a week.

CVBoost is AI-powered CV tailoring built for Cameroon's job market. Upload your CV, paste the job description, get a tailored CV and cover letter back in seconds. No prompt engineering, no settings to configure.

Why Groq's LLaMA 3.1 70B

The model runs on Groq, and inference speed is the product decision here: the user is waiting mid-application, not kicking off a batch job. The pipeline takes two inputs — the candidate's existing CV and the target job description — and returns a CV reshaped for that role plus a matching cover letter.

Language and payments, handled locally

The interface works in French and English, both directions — a job market that doesn't run on English alone. Premium use is paid through CamPay, supporting MTN MoMo and Orange Money rather than assuming everyone has a card. There's also job scraper support, configured through environment keys.

Under the hood

Auth: JWT with httpOnly cookies
Performance: response caching plus rate limiting on Redis/Upstash
Stack: Node/Express, MongoDB (Mongoose), React 19, Vite, Tailwind v4

Takeaway

Tailoring is the step people skip when they're tired — which is exactly when they're applying. Automating that one step is the whole point.

Try it live: https://cvbuilder-obfc.onrender.com

Feedback welcome from job seekers and developers alike.