VerifAI

Inspiration Scholarship and opportunity scams keep showing up in Nigeria, and not as one-off news stories. Africa Check, Dubawa, and FraudNet have each published warnings about fake scholarships in the past two years, all on their own. Fake portals copy real government domains, messages claim you've already been selected, and some scams have cost students millions of naira through fake notarized agreements. At the same time, even real opportunities are hard to judge. A student sees a link shared in a WhatsApp group and has no easy way to know if it's worth their time, whether anyone has actually won it before, or what a strong application even looks like. We wanted to build something that could actually check, not just guess. What it does You give VerifAI a link, a description, or a screenshot of an opportunity. It investigates it the way a sharp, skeptical senior would. It checks whether the organization has a real footprint online, searches for actual past winners or alumni, and verifies that the application link goes where it claims to go. Then it gives you a verdict, verified, caution, or likely fake, backed by the sources it actually found. If the opportunity holds up, VerifAI doesn't stop there. It looks at what past winners had in common and builds a short, specific guide on how to apply well. How we built it VerifAI runs on a six-step agent pipeline, not a single prompt: • Extract claims from whatever the student pasted or uploaded • Verify the organization has an independent footprint online • Search for named past winners or alumni • Verify the application and deadline links actually resolve and match the real organization • Weigh all of that evidence into one verdict • Generate an application guide from the patterns found in past winners Steps that need live information use real search and page-fetch tools instead of answering from memory, so the output can always be traced back to an actual source. The app is built on Next.js with PostgreSQL and Prisma, extending the technical foundation we'd already built for AppliQ, our earlier opportunity-tracking project. Challenges we ran into Keeping every step honest was harder than it sounds. It's easy for a model to round weak evidence up into a confident answer, so we had to be explicit that “we found nothing” and “we found evidence of fraud” are two very different things, and the verdict has to say so clearly instead of defaulting to a comfortable middle answer. Working solo on a tight timeline also meant constantly deciding what to cut. It was tempting to add more, deadline tracking, a discovery feed, reminders, but the whole point was to keep the verification loop sharp and actually finished rather than half-built and broad. Accomplishments that we're proud of Getting the agent to actually behave like an investigator instead of a chatbot. Every verdict VerifAI gives is backed by real evidence it went and found, not a single-pass opinion. Watching the live investigation happen step by step, checking the organization, searching for winners, verifying links, is honestly satisfying to watch, and it's proof the AI is doing real work, not decorating a simple lookup. We're also proud of how cheap and scalable the whole thing is. The pipeline works the same way for one opportunity or a hundred thousand, in Nigeria or anywhere else, with no manual curation needed. What we learned The biggest lesson was that meaningful AI use isn't about a bigger prompt, it's about giving the model real steps to take and real tools to use. A single well-crafted prompt can sound convincing without actually knowing anything. An agent that has to go find evidence before it can answer is a completely different, more honest kind of product. We also learned how deep the scam problem really goes once you start researching it properly. It's not a rare occurrence, it's a recurring, well-documented pattern that keeps hurting students who are just trying to get ahead. What's next for VerifAI A shared verification library is the obvious next step. Once one student checks an opportunity, the next student checking the same one should get an instant, cached answer, so the product gets faster and smarter the more people use it, at no extra cost per check. Past that, the same pipeline that verifies a scholarship can verify other things students get pitched, investment schemes, fake job offers, anything that follows the same “too good to be true” shape. And since nothing about the pipeline is tied to Nigeria specifically, it can grow to other African campuses without rebuilding anything, just local awareness and word of mouth.

Built With

• css3
• typescript