MLOps: A Leader's Perspective // Stephen Galsworthy // MLOps Coffee Sessions #39

Coffee Sessions #39 with Stephen Galsworthy of Quby, MLOps: A leader's perspective.

//Abstract

//Bio
Dr. Stephen Galsworthy is a data leader skilled at building high-performing teams and passionate about developing data-powered products with a lasting impact on users, businesses, and society.

Most recently he was the Chief Data and Product Officer at Quby, an Amsterdam-based tech company offering data-driven energy services. He oversaw its transformation from a hardware-based business to a digital organization with data and AI at its core. He put in place a central cloud-based data infrastructure and unified analytics platform to collect and take advantage of petabytes of IoT data. His team deployed real-time monitoring and energy insight services for 500k homes across Europe.

Stephen has a Master’s degree and Ph.D. in Mathematics from Oxford University and has been leading data science teams since 2011.

//Takeaways
MLOps as a process, people, and technological problem.
Experiences from a team working at the forefront of data and AI.

--------------- ✌️Connect With Us ✌️ -------------
Follow us on Twitter: @mlopscommunity

Timestamps:
[00:00] Introduction to Stephen Galsworthy
[01:28] Stephen's background in tech
[03:53] ML at scale and productionizing
[05:28] "Production is not the end."
[06:15] "I know what it means to start with nothing, to start with data, to start with trying to understand the business problems, and to make that step from zero to one."
[07:35] Technical challenges: the difficult parts aren't the technology parts
[09:13] "The technology, in comparison to 5 years ago, is not the big stumbling point anymore."
[09:37] "Allowing the data scientists and engineers you have in the organization to focus on where they can be at the most value, that's gonna be closer to your specific use cases/business problems rather than reinventing the wheel."
[10:20] "You focus on your efforts where you can make a difference to the business."
[10:30] Organizational notion of MLOps
[18:00] "From machine learning problems, a lot of the challenges in collecting label data, if you want to breakthrough in a machine learning problem, it's that label data that's gonna set you apart."
[20:43] Implications of staying at the same page with the stakeholders
[21:05] "It's really important for different levels with stakeholders and what found is that different approaches work best."
[25:34] "To be able to understand the stuff that you need to know and the stuff that you can ignore is tricky because there's so much noise around the technology. You need to be destructed a little bit more to get those strategic consequences rather than into the nitty-gritty details."
[26:54] Stephen's role in Quby, their mandate, and what they do
[28:30] "Title was always a little tricky one to stick on because it's more than just what a Chief Data Officer is doing in many organizations. It was really getting cored on what the product means, the product strategy, and also the company strategy in making that transformation."
[31:37] MLOps Organizational Challenge Project
[32:15] "I think the best example is not the first project. The first one, it always takes longer than you expect."
[35:37] "When you've got the technology in place when you've got a team and an organization that is jelling together and you've got discipline in the processes, you can move very quickly."
[37:00] "Process is like traffic lights. It might slow down an individual driver but it makes the whole traffic across the city flow a lot more smoothly."
[38:07] Impactful communication with the stakeholders
[41:34] "I think it's important to have that openness, it's sharing particularly with peers at the leadership level of what you anticipate the challenges are right now when you anticipate a payback because your first machine learning use case is always gonna cost a lot of money."
[43:25] End-user benefit crucial to getting the whole by-end.
[43:44] "Going into the outside world and talk about it helps a lot not only for getting the word out and explaining what you're doing but also getting your colleagues into the organization to understand it and become proud of it."
[46:06] Directing problems to be the biggest impact on the business
[47:05] "Probably 95% of the time the answer is not Machine Learning or certainly it shouldn't be the first thing you do. A lot of the time, a very simple data-driven approach can work very well or just completely a non-AI approach."

[zhaodandk]

This is so valuble. Thanks for sharing :)