From Chemistry to Compliance AI: How I Actually Got Here
I started university studying Chemistry. Six months in, I left. Not because the field is bad — it isn't. Because I kept being more interested in the questions behind the experiments than the experiments themselves. This is the actual story of what happened next.
The Pivot
I enrolled in Chemistry at the University of Ghana in September 2017. The choice made sense on paper: I was good at science in secondary school and Chemistry felt like a credible direction. Six months in, I realized I had made a category error. I wasn't drawn to the material itself — I was drawn to the idea of using systematic thinking to solve hard problems. The bench didn't offer that in any form that felt real to me. I withdrew before the second semester ended.
What I did next was read. Not textbooks — books about figuring out what you're actually good at versus what you think you should be good at. Two that helped were The Purpose-Driven Life and The Gift in You by Dr. Caroline Leaf. What I took from them, stripped of the framework, was a simple observation: I was drawn to systems, to programming, to the structure of problems. Computer Science had all of that. So I transferred and enrolled in CS at the University of Ghana.
Finding Data Science, Then AI
Early in my CS studies, a friend mentioned a data science program offering sponsorship toward employment. He wasn't interested — it required too much math and programming. I was immediately interested for exactly those reasons. I spent several weeks reading everything I could: how data science actually works in practice, what the job looks like, how people get in. I reached out to practitioners on LinkedIn and found a mentor, Yvette Kondoh, who gave me a grounded picture of the field that the internet's hype pieces hadn't.
That research pulled me toward the engineering end of the spectrum — less visualization and dashboards, more building things that work in production. From there, AI was a natural extension. What specifically drew me to it: the combination of math, engineering, and problems that don't have a clean answer. A classification model is never fully done. A verification pipeline can always be improved. A compliance system has to evolve with the threats it's designed to catch. That kind of ongoing, adversarial problem felt like the right kind of hard.
What the Learning Curve Actually Looked Like
Changing fields mid-university is not a smooth process. I had to rebuild a CS foundation while catching up on prerequisites I'd missed, and I was working part-time alongside my studies. The learning was non-linear — some weeks I made fast progress, some weeks I went backwards. I started with online courses and worked my way toward more hands-on projects where the feedback was immediate.
The most important thing I learned in that period was not any specific framework or technique. It was how to pick up a new one quickly when a project demands it. PyTorch, TensorFlow, LangGraph, FastAPI, Neo4j: none of these were things I knew going in. Each one was something I learned because a specific problem required it.
Where It Led
Most of my work now sits at the intersection of AI systems and financial compliance: identity verification, AML transaction monitoring, deepfake detection, PEP screening. All of it built for African markets, where the infrastructure assumptions most AI systems rely on — stable connectivity, centralized identity databases, standardized documents — don't hold.
The Chemistry-to-AI arc sounds more dramatic than it felt from inside it. What actually happened was a series of practical decisions, each one responding to what I discovered I was good at and interested in. The through-line was always problem-solving: first I looked for it in the wrong place, then I found it in the right one.
