Andrew Louw

The Apollo Program, the Bloodhound SSC and TV shows like “How it’s Made” all showed me that the world is full of problems waiting to be solved. Some of the problems are very practical and can improve lives in a very real way, and others can be added to a grander list of humanity’s achievements; as a student I found the latter much more inspirational – but since entering industry I’ve found excitement in improving efficiency in lots of small ways. This set me on a science based career path very early on, with a focus on engineering as I discovered how much I enjoyed practical work, and designing.

I knew that University places, as well as career spots, are highly competitive. I wanted something that would help make my CV stand out and I thought that Arkwright would help me with that. I found the scholarship to be especially useful when looking for work experience and early jobs - it was certainly the main factor in my first 3 weeks of unpaid work experience; including one week with a company owned by my sponsor (Pearson Engineering) and two weeks in a related field (IHC EB).

I studied Engineering at university – the course is as broad as possible in the first two years. This is a mixed blessing on the one hand you can discover new areas in engineering and transition (as I did); but on the other, you have to study Thermodynamics for 2 years, a very mathematical topic that I found difficult to intuitively follow. It was here that I first discovered control and systems engineering, and information theory, which I pursued in my third and fourth years. This pairing has strong connections to a machine learning technique called “neural networks”, a topic with which I soon became obsessed and this became the focus of my fourth year project. Machine learning has been a fast growing field over the last few years, it’s focus has been developing general algorithms that allow computers to solve problems using large amounts of data instead of human understanding. This is especially useful for problems we don’t fully understand, or where coding a solution is too time consuming and difficult.

My first job after university was with a small start-up trying to create a recommendation engine. Start-ups are always great learning experiences because they generally don’t have the time or manpower to spare you doing a pointless project (like some grad schemes I’m aware of) – as a result I found myself working on the production codebase from week 1. This forced me to learn some coding best practices, as well as quickly develop a machine learning based recommendation system, as well as many other features for the app. Recently I made a move to a much larger company, CGI, they are a multinational technology consultancy who are new to providing machine learning services in the UK. My role as a data scientist has been to help win work and grow the data science skillset within the UK. This has included making proof of concept code projects that show how sensor data can be used to predict machine failures to allow proactive maintenance, to show how customers can be clustered based on their spending habits to help give smarter promotional deals. I have also done some deeper investigations to help put together proposals for 15-month projects that will require a large amount of research and development. This has been a nice mix between short-term actions and long-term planning, This included a project based in Germany for 6 weeks but unfortunately this has been postponed for a few months.

At CGI we have split all the work associated with Machine Learning into 3 separate jobs: data engineering, data science and machine learning engineering. All three are quite specialised, but work together to produce all the features that our clients need. Data engineering is the process of setting up data stores so that data can be accessed by the people or programs that need it when they need it. I, the data scientist, take the data and perform experiments to see if it can be used to predict useful variables, often looking for some sort of signal in the data. The output of these experiments might be a piece of code that can make a prediction, but this is very different from a product our clients can use. The machine learning engineer takes that code and surrounds it with systems that allow for monitoring, scaling, redundancy and other tools to allow the final product to be highly reliable, quick and more versatile than the original proof of concept.

The field of data science is constantly growing with new software tools to help run experiments, along with new understanding about what might work and how to best use the data. Most new projects begin with a research phase so that I can understand how to approach the problem. This has been similar to my university projects and I often read academic papers for inspiration and to keep up to date with the latest theory.

Arkwright provided a great source for contacts, advice and help to progress into industry at a much earlier stage than many people are able to achieve. As a result I have found myself further progressed in my career than many of my peers. I also found the great people at the foundation to be immensely supportive when I was finding the university workload difficult. The scholarship itself allowed me to gain access to enterprise grade software for design work that I found useful for course material and to keep up my skills for my career.