CTO interview: Clive Cox, deploying the power of Machine Learning
In this interview, we’re talking to Clive Cox, CTO at Seldon. He shares how he got interested in artificial intelligence and then into machine learning. Now at Seldon, Clive helps other companies run machine learning. Clive discusses how there are many ways to get into his field nowadays, especially compared to when he joined the industry.
You’ve done prestigious studies, followed by a 25 years career. Can you share any insight into what lead you to where you are today?
I suppose I’ve always liked CS from an early age. I was using the BBC microcomputer. I enjoyed that, and it was the early days of AI, so I went to Cambridge for a MSc in Computer Speech and Language Processing. That got me in the area of speech recognition that interests me a lot. I did a PhD in that field for 5 years, and then I wanted a break from academia. I joined Logica, doing something completely different. It was a big consultancy company at the time in the space division. I worked on a satellite project for one year.
But then I got back in speech recognition, joining Vocalis in Cambridge to do speech recognition over the phone for banking apps. It was in the very early days of paying online in the tech boom. The company was reasonably small and expanded quickly to nearly 100 people. We IPO’ed but hit the tech boom and were bought by another company.
Alex, Seldon’s founder, joined a location app at the time, but it didn’t work out. When he created Seldon, I joined at the very early stages as employee number two, so I was there since the very beginning.
You were at Seldon since its creation. What were the key things you had to look for when setting up the grounds for this new project?
You really need to have the key inspiration of what will make a difference when you start. Alex had good contacts for seed investors, getting the company off the ground. We pivoted, starting from recommendations, but most companies had the pain of running large systems.
We realized many data projects in companies were not getting in production. Google did a paper showing the technical debt of ML projects. We saw companies did not concentrate on the full production lifecycle, and that’s where we saw a big gap in the market.
Speaking of, what is Seldon? How does it differentiate from big cloud providers providing Machine Learning features?
We’re not a consultancy. We’re a platform where large enterprises come to us. Companies that have large data science teams want one single place where they can run models, do auditing, and centralize all the information. It enables governance of ML models. Each team uses different techniques.
There are lots of tools on the market, like SageMaker, but they give you basics all along the line. We specialize on monitoring models with specific techniques implemented in our platform. It helps understanding those techniques and models.
Can you walk me through the tech you’re building?
We run on Kubernetes and are not tied to any cloud. We made a good choice to back Kubernetes. We chose it, and luckily, we were right. Our platform runs on that. Our clients can choose where they run them. The client installs our software in their cloud or premises. It simplifies their security concerns as they have full internal control of the data.
A lot of our projects are forward-looking, trying to see how models can run at scale. Because of Kubernetes, we use the languages and tools of Kubernetes like Go, which is the key language of Kubernetes, Prometheus, and Elastic, plus all the standard cloud native tools that go with Kubernetes. The other part is that, although we don’t create models, we have models to monitor other models, like meta-models in a way. We have lots of Python, Pytorch, etc.
We have various Open-Source projects in the area like ML Server, an optimized Python server to run ML models, to give back to the community. We also have Alibi Detect and Alibi Explain to monitor and explain models.
We have a core set of building blocks becoming standards in the industry. We have Seldon Core for example, and these Alibi projects for monitoring ML models. We build our core enterprise model on top of that.
How close does your tech team work with your clients?
They work quite closely, making it easy for them to run our product. We do have teams that help them decide what part of the system they want to run and what techniques they want to apply.
What advice to you have for anyone wanting to start a career in Machine Learning? Besides listening to Andrew Ng on Coursera!
I suppose you have to choose which area you want. There are so many areas today. We do ML operations, a cross-over between DevOps, computer science, and machine learning. You must understand whether you want to get more into operations or data science, creating models and working on core data sources.
The big decision is more to choose whether to go for an enterprise or startup. There are advantages in both to have a big impact.
If you want to connect with Clive, click here.
To learn more about Seldon, visit their website: seldon.io
If you’re a techie working on something exciting or you simply want to have a chat, get in touch with me. I’m currently CTO at Kolleno.com
FROM THE AUTHOR
>> 5 reasons your invoices are being paid late