#212 Structuring Volvo For Operational Success with Leonard Aukea, Head of Machine Learning, Engineering & Operations at Volvo

The motor industry has always been right at the forefront of innovation, and this is also true when it comes to embracing machine learning and AI. This week’s guest on the Data Futurology podcast is Leonard Aukea, the Head of Machine Learning, Engineering & Operations at Volvo, who shares with us insights into what the global vehicle giant is doing to bring value to the operations chain across the company.

For Aukea, it has been a story of establishing best processes across the organisation. He said that one of his first priorities was to bring the various data science teams together to minimise the impact of siloing, and encourage the machine learning practitioners to adopt software engineering principles. This might not be immediately comfortable to them, but as Aukea said, ML experts are smart people working on complex problems, and facilitating an open-minded approach across the organisation is key to driving long-term success.

“You need to start simple,” he said. “Think about processes, ways of working, and the cultural aspects, and try to fit tooling and infrastructure along that kind of endeavour. You don’t need to choose the most extreme state-of-the-art tools.”

At one point, Aukea noted, things being pushed into production were becoming unmanageable, so he and his teams took a step back and reset. “We went back and decided to focus on first principles,” he said. “We evangelised these first principles to develop good ways of working, and then adopted the infrastructure and tooling towards building AI on top of that.”

Ultimately, Aukea said, quality comes from the processes, rather than the technology. There are, of course, technical challenges, but for anyone aiming to get true value out of machine learning, the focus needs to be on the processes.

Aukea then explains how, with those processes in place, he and his team have been able to start delivering deep and valuable insights. For more on how Aukea was able to structure Volvo for success with machine learning in operations, tune in to the podcast!

Enjoy the show!

Thank you to our sponsor, Talent Insights Group!

Join us in Sydney for Ops World: https://www.datafuturology.com/opsworld

Join our Slack Community: https://join.slack.com/t/datafuturologycircle/shared_invite/zt-z19cq4eq-ET6O49o2uySgvQWjM6a5ng

“It’s not rocket science. These people are smart individuals working on complex problems. It’s about having an open mind trying it out. And then over some time we see that this actually becomes more or less second nature to the way they’re working.”

— Leonard Aukea, Head of Machine Learning, Engineering & Operations at Volvo

WHAT WE DISCUSSED

00:00 Introduction

03:00 Over the last couple of years, we've done polls about how many models people have organisations have in production. And in Australia, at least most organisations have zero to five models in production, with the majority having zero. Is it hard? And why is it so hard?

05:31 Is there a different approach that the data science teams need to take, to be able to create products in production and have that industrial strength to the products? What are some of the changes in mindset and approach that the teams need to make?

07:42 What are some ways you've seen organisations having better quality data or focusing on the data component. What are some ways you've seen that work?

09:30 I also wanted to ask you about the differences between CI CD pipelines. How does the data science or ML CI CD pipeline differ from the software engineering, CI CD pipeline? Are there any divergences or is it mostly overlapped?

14:24 What's the handover point between the data processing pipeline and the model?

16:51 What is the model creation component look like for you?

18:51 Do you use cube flow by any chance?

19:43 What recommendations or tips would you have for people that want to move into this space?

26:49 I've seen a couple of open-source approaches to data versioning. What do you think about those? What are your main frustrations with the way it's done at the moment?

31:27 What are your tips, recommendations, or learnings from when you were starting the journey and in further phases?

36:07 How did you make the transition and fill in the gaps and upskill?

EPISODE HIGHLIGHTS

• Securing that end to end machine learning pipeline is something that we're definitely working on, and trying to generalise that to work for the different endeavours.

• Start simple. Think about processes and ways of working and cultural aspects and try to fit tooling and infrastructure along that kind of endeavour.

• We went back to let's try to focus on first principles. Try to teach and push and evangelise these first principles so that you have good ways of working, and then adopt the infrastructure and tooling towards that and build the AI on top of that.

• If you want to secure effective experimentation and effective development, you need to be able to, or the practitioners need to be able to compare apples with apples and go back in time and see what actually was done and secure a good method for comparing. Was this approach better than the previous approach? And ideally even that is automated, when you do have good control, good metrics. And you think about those things early on in your endeavour.

• If you can show that this is valuable and tell the story when you're working with data, have some clear insights, that's where you can actually get people to make big impact decisions based on data where you establish trust and proof of value. You need to let it sink in. And you need to have the right buy in from the right people.

• It's crazy how much valuable you can become and how much more effective you can become, as a traditional data scientist, with just some good core knowledge and software engineering practices, becoming a bit better at coding and reusing.

---

At Data Futurology, we are always working to bring you use cases, new approaches and everything related to the most relevant topics in data science to help you get the most value out of these technologies! Check out our upcoming events for more amazing content.