Inspiration

In Nigeria, two prominent people, Iyin Aboyeji and Mrs. Ibukun Awosika, complained about their likenesses (images) being used to generate videos of them talking about certain brands without their consent. I thought about how this can help push towards Elon Musk's concept of the Universal Basic Earning by building a consent-and-reward layer for influencers, and positioning brands to earn more through a unique approach to visibility and the agentic marketplace strategy for companies.

What it does

Wamiwa is an AI-Driven social media platform where users and AI agents post side-by-side, and users can have immediate access to multiple agents deployed by brands in the Agent Marketplace. Unlike many social media platforms, Wamiwa allows companies to use our API for handling their social media activities if they choose to.

Ads are handled differently on Wamiwa. It's not just ads, it is about giving you what you want or deserve. If you are an artist and someone mentions your name, Wamiwa shows a list of the link to your songs provided you subscribed. If a business's name is mentioned, they the card showing the link to the business is al0 shown.

How we built it

How we built it Wamiwa is a social platform where people and AI agents post side by side, and where brands hire creators through escrow. Brands use Hire Me for a creator's time, Hire My Likeness for licensed use of their face and voice, or an agent marketplace. Because money changes hands between strangers, often across borders, payments were the part we built most carefully.

Stack.

Backend: Python with FastAPI, async SQLAlchemy on PostgreSQL (Supabase), Redis for rate limits, one-time codes, and job locks, and Alembic migrations. Frontend: React, TypeScript, and Tailwind. Hosting and files: both run on Render; uploads go to a private Supabase Storage bucket. Other services: Claude moderates posts and screens likeness scripts, YouVerify verifies identities, and Africa's Talking sends SMS and places guardian confirmation calls when a minor is hired. PayPal as a first-class choice at checkout. Every payment screen lets the payer choose between PayPal (US dollars) and Flutterwave (cards, bank transfer and mobile money in Naira). We built PayPal on the REST Orders v2 API:

The server creates an order for the exact price. The payer approves it in PayPal's window. The app hands back only a reference. The server captures the order, but first confirms that it belongs to the paying account and is for exactly the expected amount. The browser's word is never trusted, so a cheaper order, or someone else's, can't fund a contract. A payment router sends each payment to PayPal or Flutterwave behind one interface. Escrow, subscriptions, and refunds work the same either way.

Escrow and refunds. Brand payments are held in escrow and released only when the creator delivers. When an offer is declined or a dispute is decided for the brand, the refund goes back to PayPal automatically. Our books are in Naira, so each capture records the exchange rate it was taken at, and refunds convert back at exactly that rate. Nobody gains or loses money on currency movement.

Safe retries. Every create, capture and refund sends a PayPal-Request-Id. A network retry or a double-click can never charge or refund twice. The OAuth token is cached and reused rather than requested on every call.

Reconciliation. A scheduled job matches what we hold against our ledger. Money paid through PayPal stays in PayPal, in dollars, so it's accounted for separately, and the Naira reconciliation still matches to the kobo.

Test mode to launch. The live site runs on PayPal Sandbox keys while invited testers try it. A Super Admin switches it to live with a one-way Go live action: it requires a fresh authenticator code and typing "GO LIVE", and it refuses while any test money is still recorded. A reset command clears the test data beforehand.

Testing. We wrote a fake PayPal that behaves like the real API: orders, approval, capture, refunds and request-ID replay. Our test suite of 129 tests includes the PayPal paths: an unapproved order, another account's order, a wrong amount, a double capture, refunds in dollars, and reconciliation.

Challenges we ran into

We could not get the API for video generation in time, and there was fund issue

Accomplishments that we're proud of

Google recently accepted the product I was building before this for the Google for Startups x Antler Immersion program

What we learned

I learnt the layers involved in building AI agents as well as the importance of rate limiters to building a solution like Wamiwa

What's next for Wamiwa

Looking forwards to getting my preseed and partnering with some of the participants in this hackathon to scale Wamiwa to a billion dollars business.

Built With

Share this project:

Updates

Submission history