In the era of Sha G BT, I think applications of A I go far beyond complex conversational bots. I would be curious if you are curious actually of the applications that you could enterprise, you could apply in your enterprise. And more importantly, maybe how you could leverage tools as well as the public cloud or maybe tools on the public cloud.

I'm Andrea Montano, ML, OS product manager at Canonical, the publisher of Ubuntu. And today I'll talk about as you might guess open source MLs platforms on AWS or on the public cloud.

Of course, applications of A Io in are available in all industries from card fraud detection in financial services to turn prevention in retail or drill summarization in oil and gas. And that's exciting. But the main question is how do we scale it up? How do we actually help organizations depending on their journey and where they are, how ready they are to implement A I to actually make it happen.

But I started my own journey here. Uh and actually how I got introduced to the public cloud for a I project. A couple of years ago, I was interviewing someone to join our company and they told me about a project that they had, they use crayfish or they, their parents, his parents had a crayfish farm and I'm not sure how much you know about it. I didn't know much. But the truth is that when they lose their shell crayfishes, they can die if the shell is not removed right away. It translated into the family spending a lot of time just monitoring all these buckets for crayfish grows. So the the student back then he was a graduate actually build a system, build a model that would run on surveillance cameras that would alert automatically the family when the shell of the crayfish is lost, so they can go and remove it. It optimize their operations.

However, he wanted to scale his project and help other companies. That's when he got, he, he moved his open source tools and his, his project on AWS. Well, that sounds easy. But the truth is that when it comes to A I, not, everything is so easy.

I like to classify the challenges in four big buckets and about the skill gap on the market. I think everyone here knows about it. We don't have enough data scientists. We don't know how to maintain an ML ops platform and we struggle actually hiring. But then we look at the operations and maintenance and scalability are the questions or the issues that we, we hear the most about and when it comes to technology, I think you can read in the news every other day about a new tool, a new platform, a new framework that's popping up. How do you choose the right one? How do you choose the one that's the most suitable for your organization? That's where we come in and help. But then also data is the heart of ne a project. If we think of a bank or at the healthcare indust uh a health care organization, they often work with highly sensitive data. But more importantly, the stats are saying how the volume of data is exponentially growing year on year. All these to put together require more compute power to address as well.

Now I like to say that everyone uses A I nowadays you have it in all industries. But then the truth is that also companies are in different stages in their A I journey, some of them are just getting started because cha gp t challenge them to rethink their A I strategy where some others are moving away from legacy tools and that's fine.

But let's not make it any longer. Let's look at some use cases, smart cities. Well, I live in Dubai and they are building a smart city there. I know some others here in the US as well. And the truth is that A I is the heart or the core of all the smart cities projects from traffic optimization to solar panel optimization. There are tons of applications and about gen A I and how it helps everyone. Well, we all know it. If I think of chatbots or image generation, there are plenty of applications already open source or not. And the media and entertainment industry or the gaming industry here also benefits from A I but they all seem as being applications for large enterprises.

But let's make it simple. Actually, artificial intelligence helps all of us. It makes our lives easier because it leads to less pollution. If you think of the street light uh uh sensors, it helps us have predictable costs and it offers us better experiences. Just things of all the system recommendor that help you out when you buy things.

But what the open source role, you might wonder. Well, in Canonical, we have the open source in our DNA. So of course, I'm biased, but at the same time, I've seen an entire movement and shift in the A I industry where everything is open source. So most of it, it's open source and the truth is that it's available to everyone. It's easy to get started and also it's easy to contribute to add new features. Do you want something that is not available in a tool? There is nothing to stop you to write some code. And more importantly, someone else might need it as well and maybe it's not that important initially, but having something that's free to use, especially for early adopters for students. It's incredible

Now when it comes to Canonical and our ML solution, maybe it's important to think of how you can plug it in and plug it out depending on different tools because a a projects are like innovation from the past. They there's they don't benefit from an out of the box solution, each industry, each company, and more importantly, each use case is different. So having the freedom of choice as well as being able to optimize your costs is crucial.

Since we also mentioned about the journey that you take on A I, it's worth mentioning that it starts with the enterprise readiness. Most of the times you start on a laptop, if it turns ubuntu, it's, it's amazing if it doesn't, maybe you should think of it. But then as the project grows and matures new tools come in place and that's when it can get complicated.

But how do you get started on the buntu machine? That's what we like. But then usually you get started with simple tools. Jupiter notebooks are the favorites of all data scientists and probably ML flow for experiment tracking. I guess that's the usual setup that I've seen.

How do you scale? However, is the main question and when companies start scaling their projects, most of the times they want faster time to market, it often translates actually in faster training of the models, a bigger teams to to run their project so user management is needed as well, but also more compute power. That's when we advise customers to leverage what they have already if it's on prem or public cloud. But also think beyond it, use hybrid or multi cloud scenarios depending on how it's suitable and train and use your data based on it. Optimize the usage of GP Us, optimize your networks as well.

Freedom of choice. It's another question that we often get and it's not just about the tools but also about the projects and the ability to migrate from one environment to another. You might guess.

Now we of course, we advise people to use open source tools, the skill gap i already mentioned about it. And it's, it's really important because as long as you don't have enough people to run your project, you might feel overwhelmed as a company. However, you can, you can help the team that you already have a lot of the work in the data science world is done manually. What if you would automate all that work such that they could spend time on meaningful activities? What if you would use MS platforms that actually enable them to focus on building models rather than integrating tools and dealing with tools, compatibility because that's frustrating, i have to say and security and compliance.

I already mentioned the fact that many A A projects run on highly sensitive data and that's where questions come in place. How do you ensure that your data will not be stolen. How do you ensure that your project will not be maliciously used or taken in any way security patching. It's one of the things and if you've heard about by shell from a couple of weeks or months ago, that's when you start looking, uh tools that are distributed by a trusted vendor.

And the truth is that you can run A I on any cloud. I've heard companies being able to run on public cloud and focusing on that. I've heard companies running on prem, which, which choice to have or w which, which option is more is better for you. I would say it depends, but more importantly, don't think of the cloud when you start your A project, think of the problem that you have.

Don't do A I because your company or your manager told you about that because your competitor does that because you've seen in the news about charge GP T it's all nice. But does it really solve a problem for you? Do you really need gen A I in your company or maybe you just need some predictive analytics which they have been there already and there is nothing to stop you benefiting from them. And that's the first thing that you should bear in mind, run A I only if you have a problem to solve.

And after that, look at the data, what data do you have? And actually do you have enough data, where does it live? Only after that we go to that to the cloud part and how to leverage it optimize based on what on the infrastructure that you have in place. If you mostly use AWS. Perfect, get started there. And after that, build a long term strategy, but if you have GP U on prem, you should leverage them and you should ensure that you optimize the usage of them because nowadays, many companies have GP Us and they utilize them around 20 to 30%.

After that, build a long term strategy and ensure you have plat a platform that enables teams to collaborate, they can reproduce their work and they don't duplicate a lot of the things that they do. And that's probably one of the most important lessons that i would like us to leave this room with.

And before we finalize, I'll go back to the crayfish story. I love hearing from our community and from people in general, what projects do they work on? How do they benefit and how do they leverage A I? That's what we love the most.

However, I would like to ask you, what's your A, A project or your crayfish project that you could get started on your own laptop and then scale on the public cloud if that's what you want. Because if you identify that, then probably you could help your team, your colleagues, maybe your home. I don't know, maybe you have some fun projects that you can work on in your own home.

And that's probably where I'll soon wrap it up, use open source tools because they are free to use and easy to use. Of course, and don't shy away from contributing to them platforms such as Q flow ML flow Jupiter notebooks Spark are a couple of the famous ones and they've been growing just because people try them out and try to improve them based on their needs.

Bear in mind that if you bump into a problem, it's very likely that you are not the first one, just google it, go to the slack channels and you'll find the solution. And don't forget, you can always innovate at speed with open source A I.

And if you would like to learn more, we do have a podcast. It's called Ubuntu A I as well as a publication. We mainly talk about SMIs open source MLs and open source A I. We have different guests and also if you would like to talk more to us, we are at boots 301 on the diagonal uh of this place to hear more about. And I think we also have time for some questions. I'm not sure if you have a mic, but if you do have questions.