In this role, I lead the organization's Agentic AI initiatives and shape how AI is adopted across the enterprise. I run a team that develops AI solutions, and I look for gaps in the business that could benefit from AI powered solutions, then work with those departments and my team to make sure an effective solution is designed, delivered and adopted.

My primary focus is building a scalable enterprise framework that enables the organization to deploy AI agents securely, responsibly and effectively. The framework sets out how agents are built, governed, secured and run, so that teams across the business can adopt AI with confidence, at a predictable cost, and at scale.

I engage the business to understand their processes and pain points, identify the gaps where an AI powered solution would add the most value, and translate those into a solution design together with my team. I then run the delivery of those solutions end to end, working with the relevant departments through to an implementation that is adopted and measurably effective. This is where cutting edge AI innovation is bridged with real business impact, empowering teams to work smarter and unlock new levels of productivity.

In this fast paced role, I hit the ground running, working on several projects in parallel, some of which have an iteration already in production. A highlight of the projects in question are, without exhausting the full list, as below:

The biggest project, spanning across three departments in the organization. The task was to implement machine learning based solutions to help with detecting Fraud, potential money laundering activities and patterns, as well as potential gambling abuse behavior. The business had a lot of data but none of it was prepared for machine learning, and there was no already prepared feature store to train algorithms on or ground truth from which training features for supervised learning can be built.

The strategy employed was to utilize unsupervised learning first, utilizing algorithms like K-Means clustering, Isolation Forest and Gaussian Mixture models to categorize gamers into 3 categories of "high risk", "low risk" and "medium risk" in terms of Fraud, AML and Responsible gambling. We then relied on business users to review, through a web portal we built, the high risk cases and tag them accordingly. This way we would be building our own ground truth for use in supervised learning later.

Through the project, we worked closely with the data engineering team, and business users to craft out new features, and utilize them for model training and prediction.

On this project, we built a Retrieval Augmented Generation (RAG) interface that users could utilize to interact with business knowledge via a chatbot like user interface. The business knowledge was integrated from various sources such as SharePoint, OneDrive and Local Disk. For SharePoint and one drive, we implemented a webhook that would listen for file changes and update the vector store accordingly so that the latest, most accurate and most recent information was made available to the chatbot users in near real time.

On this project, the business was struggling with expense capturing, claims processing and record keeping of expenses. They had been considering buying an already existing system. We then utilized our AI product solutioning skills to craft up a solution in-house.

We utilized Azure Document Intelligence studio to train a model that can extract key fields from receipts. We then built a NextJs and Python FastAPI Web solution around this model where users can either open their camera and capture a receipt or upload an already captured receipt from their file storage, which will then be parsed by the model and key information extracted and persisted in a Fabric Lakehouse, along with the URL of the blob which we utilized to display in the application when users would review their expenses. We then built a Power BI dashboard on top of the data for ease of analytics.

On this project, I was part of a 6-person team to do development work in C# .NET, python, T-SQL, Azure and Microsoft Fabric, React with NextJs.

The project had multiple sub projects in it, including but not limited to:

During this sub-project, we developed a solution to automate the client's Sox Audit process. We utilised Azure Logic Apps to orchestrate the audit process. The logic App was powered by Azure function written in C# .Net as well as SQL stored procedures on-prem that we connected to via an On-Premises gateway.

During this project, after we had achieved a stable product, I created Infrastructure as Code for our client, using ARM (Azure Resource Manager) templates so we can easily reproduce the infrastructure in any new environment.

This later proved to be useful as other companies under the same group as our client later needed the product, and using IaC, it was quick and easy to get the product up and running in their environments.

On these sub-projects, the client had a lot of their business logic in T-SQL stored procedures, and some of the procedures were becoming bottlenecks for their business processes. We were then tasked with going through the stored procedures, documenting any inefficiencies and suggesting optimisations to make the procedures more efficient in execution.

Some of the stored procedures mentioned above were also part of a bigger ETL process that ran on-premises. Such ETLs were orchestrated by means of SSIS packages. On such projects, were tasked with extending the functionality of the ETL process to add new functionality and re-deploy the SSIS packages to the respective servers.

As part of their migration from On-Premises to cloud with Microsoft Fabric, on of the sub-projects we did for the client was to automate their ETL processes using Fabric Data pipelines.

Most of the ETL processes involved integrating with third party REST APIs, getting data from the APIs, performing some transformations and aggregations, and finally storing them in either a Lakehouse or a Warehouse or both. We utilised Python Notebooks to perform this logic with Pyspark, and utilised Data pipelines to orchestrate and schedule the ETL processes.

For most of the data we imported into Fabric, the end goal was a PowerBI report that business users could use to draw meaningful business insights and make key business decisions from these Insights.

We performed dimensional modelling on the raw data we imported, to convert it into facts and dimensions following the star schema, creating meaningful measures and computed columns in the dimensional model, then used the dimensional model to create powerful PowerBI reports, with detailed and intuitive visuals for ease of interpretation for the end business user.

I ran and managed the delivery of a few of these processes with a small sub-team under me, to see the process from end to end, thus from ETL to insights.

On this project, our team was responsible for the platform and DevOps facilitation and support for Standard Bank business banking teams. We created and maintained pipelines, AWS cloud infrastructure, managed and maintained the Kubernetes clusters where the applications were deployed across multiple environments such as Development, Test and Production. We configured and provided monitoring and observability functionalities for the applications running in the cluster and supported developer teams with deployments to production as well as troubleshooting any cloud/DevOps related issues they encountered.

On the Aluminserv project, I was part of a five-person team, with three of the members responsible for the front-end, and myself and the tech-lead responsible for the back end. I was responsible for relational database design and implementation, writing and exposing APIs to interact with the stored data, and supporting the front-end developer experience as they used the APIs.

On the TransUnion project, I was part of a two-person team where we built some APIs for the client to enable new processes and services that they introduced to modernise their product offering. I was actively involved in the entire development process from requirements gathering, to system design, development, testing, deployment, maintenance, and support of the APIs. We split the APIs to be built between the two team members, with each member fully responsible for building a set of APIs from start to production, an enormous responsibility for each of the members.

On the cloud team, we built, configured, and secured cloud architectures for various esteemed clients in the financial services and retail sectors. I made use of Terraform as the IaC (Infrastructure as Code) tool of choice to carry out my tasks in building the desired architecture. We extensively made use of Terraform Cloud to automate and continuously deliver our builds to the client's AWS account.