March 20, 2025
From Applied Mathematics to MLOps Engineering: My Journey Through AI, FinTech, and Retail
As an Applied Mathematics lecturer with a passion for mathematical modeling, I never anticipated the incredibly rewarding journey that lay ahead of me. Driven by curiosity and a desire to tackle real-world challenges, my path led me to Stellenbosch, where I joined a Machine Learning startup. While I initially aspired to become a data scientist (I have a Masters focused on mathematical modeling and wanted to do modeling for a living), I found my calling in MLOps Engineering, where I could leverage my natural inclination for automation.
Initially, it was a murky, steep learning experience. But what made it worthwhile were a few key personality traits I am privileged to possess: perseverance and hyperfocus.
Academic Foundations in Mathematical Thinking
My journey began in the structured world of academia. Before studying Mathematics, I was enrolled in Mechanical Engineering. It was occasionally enjoyable, but mostly tedious. Then I walked into a second-year Differential Equations class where the professor started writing equations for Newton's Law of Cooling. I was astounded. My mind exploded. We can do that? We can describe physical systems in such a beautiful way mathematically?
To be frank, I never really knew what I wanted to do in life. I loved anything science-related and wanted to study medicine—a feeling that still returns occasionally. But I'm glad I didn't—it's not as glamorous as Netflix portrays. So I pursued Mathematics and Computer Science, much to my parents' dismay. I recall my grandfather, an engineer, asking what I planned to do with a Mathematics degree. My answer then was to become a lecturer, to which he responded, "I guess that's also a job."
Fast forward a few years. I received an opportunity to pursue a Masters with research in Mathematical Modeling—specifically, Mathematical Epidemiology. My research examined how pathogens propagate through populations via direct and environmental transmission mechanisms.
With my Master's degree from North-West University, I developed a robust framework for analytical thinking. This research established thought patterns that would prove invaluable when working with complex systems later in my career.
As a lecturer teaching Pure and Applied Mathematics to undergraduates, I honed my ability to communicate complex concepts clearly—a skill that would later serve me well when bridging the gap between technical implementation and business value in my engineering roles, or so I believe. My peers might disagree.
Early Career Breakthrough: Cloud Migration in Banking
My first significant project outside academia represented a genuine technological leap. I had the remarkable opportunity to migrate an on-premise system for training a large recurrent neural network to a cloud-based framework on Microsoft Azure. This was revolutionary at the time, as most South African financial institutions (and most across the world, honestly) hadn't yet embraced cloud computing or machine learning models.
The project presented numerous challenges—from security concerns to ensuring consistent performance and optimizing costs—but the results were transformative. And I'm not primarily concerned about the transformation at the company where I worked. This experience opened my eyes to the immense potential of technology in reshaping traditional industries. More importantly, it kindled my passion for making sophisticated technology accessible through well-designed systems and automation. I worked hard and late, learning from my own mistakes, stressed by breaking components deployed in production settings because I was so new to cloud computing. It was 2016, after all.
Then something happened that I dreaded: I broke something in production. Me. It was my fault. The guilt nearly drove me to a panic attack, and then I realized—after all the times I had broken things, experimented, and gotten my hands dirty—I was now skilled enough to know how to fix this. That changed everything. I was freed from my professional self-doubt. For a few weeks, at least.
The FinTech Years: Building Systems That Scale
Praelexis: Banking Innovation
At Praelexis, I spent five formative years developing machine learning solutions for both retail banking (Capitec) and private banking (Investec). The diversity of projects—from transaction labeling tools to cognitive banking platforms—provided a comprehensive education in applying AI to financial services.
One of my most impactful projects was a Transaction Labelling Tool that utilized transaction embeddings (using BERT—while transformer architectures were brand new) and clustering algorithms to label massive amounts of transactions on bank statements. What previously took hours could now be accomplished in seconds, dramatically improving operational efficiency.
Jumo: Financial Inclusion at Scale
My journey led me to Jumo, a dynamic company passionate about improving access to financial services through accessible credit across Africa. I joined the Prediction Services team, where I implemented automated model deployment pipelines, developed scalable training services, and established robust model monitoring systems.
The technical challenges were considerable. As a (novice) Product Owner for the Prediction Services Team, I led MLOps initiatives and fostered collaboration with the Decision Science Team. Through deductive reasoning, I later developed a set of champion model architectures that could be generalized across various modeling problems—an approach with the potential to dramatically improve our efficiency.
Revolutionary Automation
The results of our automation efforts were remarkable:
-
Automated Model Deployment Pipeline: We reduced deployment time from several days to mere hours, creating a system that could handle multiple simultaneous deployments.
-
Scalable Training Service: Our data scientists went from training a few models per week to up to 50 per day, exponentially increasing our experimental capacity.
-
Real-time Model Monitoring: We built a monitoring pipeline that detected drift in predictions and features almost immediately, rather than weeks later when problems had already affected customers.
Through automation, we reduced deployment times from days to hours and dramatically increased our model production capacity. Later, I moved to the Data Science and Analytics team, building innovative tools to enhance model training and improve predictions on customer default rates.
The Power of AI in Financial Inclusion
The impact of our work at Jumo was profound. Our models determined customers' probability of defaulting on loans, enabling responsible lending decisions. We were able to serve those with relatively sparse financial data, striking a delicate balance between helping customers and maintaining sustainable operations.
In underserved African communities, data collection presented significant challenges. We had to creatively utilize available data to accurately map customers' financial landscapes. Our partnerships with Mobile Network Operators proved invaluable, providing insights into customer behavior in informal economic sectors.
This work wasn't just technically interesting—it was meaningful. Every improvement in our models or systems meant more people gaining access to financial services they desperately needed. The technology we built extended credit and savings facilities to communities previously ignored by traditional banking, fostering economic growth in underdeveloped areas.
At the heart of our mission lay a dedication to automation and personalization. We built multifaceted systems balancing these two elements—automation allowed us to scale effectively, while personalization ensured customers received tailored services. As a relatively small team, we served millions of customers, demonstrating the remarkable power of AI to reach underserved communities and deliver previously unimaginable services.
The scale was staggering: a handful of engineers supporting models that made millions of credit decisions daily across multiple markets. This wouldn't have been possible without our relentless focus on automation and robust systems design.
Not every project unfolded according to plan. While working on a cloud-based credit prediction platform, my earnest efforts to enhance the process didn't yield the expected results. This experience taught me the importance of perseverance and effective expectation management. Each setback presented an opportunity for growth and learning.
I learned to embrace failure as a natural part of innovation. When our first attempt at an automated monitoring system struggled with the volume of data, we didn't abandon the concept—we reimagined it from first principles, resulting in a more robust and scalable solution. Therefore, it's paramount for anyone considering a career in Machine Learning Engineering to embrace failure. Without this mindset, you risk finding yourself on a poor career trajectory.
The Ethics of AI in Finance
Looking forward, AI Ethics and Regulation became increasingly important in our work as MLEs. The potential of AI to revolutionize lives is undeniable, but we remain acutely aware of potential misuse. Ensuring model explainability and their equitable service to all individuals aligns perfectly with the objective of improving lives in every aspect of AI/ML's daily use.
The ethical dimension adds complexity to our work as MLEs, but also meaning. We aren't just building efficient systems; we should also focus on building responsible ones that withstand scrutiny and earn users' and regulatory trust.
Transitioning to Retail: New Horizons at John Lewis Partnership
In February 2024, I embarked on a new chapter in my career, joining Equal Experts as a Machine Learning Engineering consultant. My current engagement with the John Lewis Partnership in the UK represents an exciting pivot from FinTech to Retail, bringing ML and AI capabilities to an entirely different industry.
At John Lewis, I've joined the newly formed MLOps Platform team, where we've taken ownership of previously deployed platforms through careful analysis and reverse engineering. While managing the Data Science/MLOps platform, we've also ventured into Generative AI use cases, and currently we're creating services that will allow Partners to deploy ML/AI systems easily and efficiently.
The retail context presents unique challenges and opportunities. The data is different, the business goals are different, but the core principles of solid MLOps remain the same: automation, reproducibility, and building platforms that empower teams to deliver value quickly and reliably.
One of our early successes has been implementing Generative AI use cases in just a few weeks, demonstrating the agility and value that well-designed AI systems can bring to traditional retail businesses. We're now focused on building a reproducible system for Machine Learning Products and development, automating the Machine Learning Development Lifecycle to enable innovation across the entire enterprise.
But one thing still holds true: The only way through this journey is with perseverance and hyperfocus.
An Afterthought: The African FinTech Revolution and Beyond
Africa experienced substantial growth in the FinTech industry during my time there, with innovations emerging rapidly. As Machine Learning Engineers, we played (and obviously still play) a pivotal role in applying AI to extend financial services to previously underserved communities. This transformation changes countless lives, and being part of it was truly inspiring.
Now, in the retail sector, I see similar opportunities for transformation. The principles of applying AI to solve real business problems, delivering personalized experiences at scale, and building platforms that enable innovation remain just as relevant in this new context.
Conclusion: Building Bridges Between Technology and Impact
My journey from applied mathematics to MLOps engineering across financial services and retail has been one of continuous learning, growth, and meaningful impact. The challenges I've faced—from personal development to hardcore business problems like data collection difficulties, modeling complexity, and ethical considerations—have fueled innovation and strengthened my commitment to building systems that make a difference.
Throughout this journey, I've maintained a passion for building bridges: between academic theory and practical application, between balancing life and work, between technical possibility and business value, between automation and personalization.
These bridges enable organizations to harness the power of AI in ways that are both scalable and meaningful. I look forward to writing more about my personal experiences and growth in future posts.
The road ahead is long, but every step brings us closer to making advanced AI capabilities accessible across industries. As I continue my work at John Lewis Partnership along with Equal Experts, I remain excited about the potential of well-designed MLOps platforms to enable innovation and deliver value in unexpected ways.