ApexScore

Inspiration

When I first saw the LMA EDGE advert on Devpost, I immediately reached out to my team. We didn’t have a clear idea yet—we just knew this was an opportunity to build something meaningful, so we started asking ourselves: what real problem can we solve?

As we brainstormed, a personal experience came back to me. A friend of mine once took a loan from a bank. It took quite a long time before the loan was fully repaid, and that situation made me think deeply. I started wondering: how do banks actually decide who to trust with loans? What systems do they use to check someone’s loan history and assess risk?

Around that same period, I came across a news report that said Nigerian commercial banks lost about $29.8 million in the second quarter of 2024 due to loan defaults. That number really stood out to me. It made the problem feel bigger than just one person’s experience, it was clearly a systemic issue.

That’s when a question kept coming back to me: why are financial institutions still relying on limited and outdated credit checks?

From that question, our idea was born. We imagined an intelligent AI system that could:

Look beyond traditional credit bureau data

Analyze borrower behavior and patterns more deeply

Support loan officers in making better, faster, and more confident lending decisions

That idea became the foundation of the solution we’re building today.

What it does

Right now, many banks and lending companies are forced to rely on limited credit data. That means they don’t always see the full story behind a borrower. Some people who can actually repay loans get rejected, while others who are high-risk get approved.

ApexScore changes that.

It brings together different data points to build a clearer picture of a borrower, looking beyond just traditional credit scores. By analyzing past loan behavior, patterns, and inconsistencies, ApexScore helps lenders understand who is truly credit-worthy and who is risky.

With ApexScore, loan officers can:

Spot risky borrowers earlier

Detect loan stacking and identity issues

Make faster, more confident approval decisions

Reduce loan defaults and losses

In simple terms, ApexScore helps lenders lend smarter, not blindly.

Impact and metrics

Using simulated applicant data, ApexScore was able to identify high-risk borrowers about 30% more accurately than traditional credit checks. It also detected potential cases of loan stacking that would have been missed otherwise and gave loan officers practical, real-time recommendations.

Even in this simulated environment, the system showed that lenders could significantly reduce defaults and non-performing loans, while approving more qualified applicants who might have been rejected using conventional methods.

How we built it

ApexScore – Technical Overview ApexScore is a credit risk assessment platform designed for financial institutions to evaluate loan applicants using both behavioral analytics and traditional financial data.

Architecture The system follows a three-tier architecture:

1)Frontend (React SPA) – Built with React 18, Vite, and TypeScript. Uses Tailwind CSS with shadcn/ui components for a polished fintech aesthetic. The app is fully responsive with dedicated mobile optimizations.

2)Cloud Backend (PostgreSQL + Edge Functions) – Handles user authentication (email/password and Google OAuth), user profiles, and search history persistence. Row-Level Security ensures data isolation per user. Edge functions handle operations like password recovery.

3)Risk Engine (FastAPI on Render) – The Python backend does the heavy lifting:

• Generates synthetic applicant data for testing
• Calculates the Behavioral Stability Index (BSI) from signals like device fingerprinting, SIM stability, and geolocation consistency
• Computes the final ApexScore using: (BSI Average × 0.6) + (TFD Score × 0.4)
• Produces lending recommendations with suggested amounts, interest rates, and reasoning

Key Technical Decisions:

• Regex email extraction in the chat interface for flexible user input
• localStorage persistence for chat messages with database persistence for search history
• jsPDF for client-side PDF generation (no server round-trip)
• Full-screen dialogs forced via 100vw/100dvh to ensure consistent experience across devices
• Multi-currency support (NGN, USD, GBP) with country-specific financial institutions ## Challenges we ran into One of our biggest challenges was access to real financial APIs. Financial institutions don’t easily give out their data, especially to early-stage teams like ours. Because of that, we couldn’t plug our AI system into real-world data.

Instead of stopping, my team and I created a mock version of the API that closely simulates how real financial data would behave. This allowed us to test our ideas, build the system logic, and demonstrate how ApexScore would work in a real environment.

Adding new features was another major challenge. I still remember what the first version of ApexScore looked like—it honestly wasn’t something we could confidently put out. So we kept building.

Day after day, we continued developing the AI system—adding features that made sense, removing ones that didn’t, and constantly refining how everything worked. It was an ongoing process of trial, feedback, and improvement.

All of this happened while we were still juggling a busy university schedule. We had to push ourselves—stretching our time, sacrificing sleep, and working through many long nights. Some nights were spent coding, others researching and rethinking the system.

Those sleepless nights slowly paid off. Step by step, ApexScore evolved into the solution it is today.

What we learned

Working on ApexScore taught us a lot—especially the importance of continuous improvement. We learned that we don’t always have to be revolutionary from the start, and we don’t need to stick rigidly to one idea. What really matters is the willingness to keep refining, adjusting, and improving as we learn.

By constantly revisiting and improving our ideas, ApexScore grew stronger over time. That process showed us that meaningful solutions aren’t built in one moment—they’re shaped through steady improvement, so they can stand the test of time.

What's next for ApexScore

After building ApexScore, we came to an important realization: this was more than just a hackathon project. We saw its potential to become a real, long-term solution.

Because of ApexScore’s ability to tackle a major problem in fintech, we’ve decided to take it beyond this hackathon and build a startup around it. We truly believe it can make a meaningful impact if developed and deployed in the real world, In the the few month to come we believe plan to secure more funds, partnership with some financial institutional bodies just to access real-time financial data, we hope to have build a better and a product that can survive in the real word.

Built With

• google-oauth
• jsfpdf
• postgresql
• python
• react
• regex
• tailwind-css
• typescript
• vite