When the Numbers Didn’t Lie — And How I Learned to Listen to the Data

The hardest part of building Arise & Shine Transporters wasn’t the code — it was the numbers. For weeks, the fuel logs didn’t add up. Drivers were reporting fuel costs that didn’t match the GPS data, and the system couldn’t reconcile the discrepancies. I spent hours chasing down every possible error, only to find that the real issue was a misunderstanding in how fuel was being measured. It was a humbling moment — a reminder that the numbers don’t lie, and that sometimes, the most straightforward solutions are the hardest to see.

What surprised me was how quickly the AI copilot feature became the most used tool in the platform. Initially, I thought it would be a nice-to-have — a way to generate daily insights for the admin. But in practice, the AI ops copilot became the go-to tool for making decisions. It flagged route deviations, suggested cost-saving opportunities, and even predicted truck performance based on historical data. It was a moment of clarity: the AI wasn’t just an add-on — it was the reason the platform worked at all.

I’m proud of the work on Local Dialect, the language learning app for Kenyan and African languages. Building it meant grappling with the technical challenges of low-resource languages, where there’s limited training data and fewer tools. I had to rely heavily on AI collaboration to design the user flow and structure the lessons in a way that felt natural and engaging. The result was an app that didn’t just teach grammar — it brought language back into the everyday lives of learners.

With Mwalimu Cosmetics, I learned how much the right data can transform a product. Early on, I assumed the e-commerce store would be straightforward — a product page, a shopping cart, and a checkout. But as the AI helped me analyze user behavior, I saw patterns I hadn’t considered. People were browsing but not buying, and the data pointed to a lack of trust in the payment process. That insight led to a redesign of the checkout flow, making it more transparent and user-friendly. It was a reminder that even the most polished code can fail if it doesn’t align with what users truly need.

Looking back, what I’ve learned is that building AI-powered tools isn’t about making perfect code — it’s about listening to the data, the users, and the numbers that tell the real story. Sometimes the hardest part is not writing the code, but understanding what it means. And sometimes, the most surprising part is realizing that the AI isn’t just a tool — it’s a collaborator, a guide, and a mirror that reflects what we might have missed.

I’m not sure what the next project will bring, but I know one thing: I’ll be listening to the numbers — and to the people who use them.