Community Evolution and the Rise of MLOps with Fred Simon at JFrog swampUP 2024

This is Techron tv. Hi everyone. It's Alan Shimel back here in Austin for our Swamp Up coverage.

This is the 10th annual Swamp Up, so we're glad to be here for that. As usual, swamp Up is an event for the community here at Jfr and what a community it is. It's been a great two days.

We're in day two. I'm really happy to introduce you to my friend Fred Simon. Fred is the co-founder or one of the co-founders Yes.

Of the three co-founders and Chief data scientist. Yes. At Jfr.

Fred, always good to see you again. Thank you. I see you've got your cowboy hat.

Yes. Are you enjoying Austin? A Lot?

A lot. Yeah. I was there for the Eclipse in April And That's true.

I remember talking to you here for the eclipse. How was it? Was it good here that day?

Yes, It was really good. Yes, there was some clouds, but we managed to escape. It was an amazing time here in Austin, and, uh, happy to be back Happy.

So we are gonna talk a little MLOps, we're gonna talk about some of the announcements here, but as a co-founder Right. 10 years of doing this, it has to give you tremendous sense of, of joy. Yes.

Right. To see how this has grown. To see, I mean, just, just a tremendous community.

It's smart people who are engaged Yes. Who Enjoy being here. Right.

How, how does it make you feel? Um, yeah, I mean, very, very good. I always enjoy talking to, uh, and you know, the customer, we know them for more than 10 years, for more than a decade.

And they coming back and every time to see them and what they are doing changed in the 10 years, what we started doing with them and what we are doing with them today. And yeah. Well, think back 10 Is So forget 10 years.

Think back 16 years to when you started, right? Yeah, Exactly. So now, So The world changed.

Never boring. Never. No.

That's what, and that, look, that's one of the reasons I think people like us are in tech. It is never boring. Yes.

If you think it's boring, find another thing to do. Yes. Um, And, Uh, but the other thing is, and and I, I don't want to give out any secret recipes, but what you will see in next year's Swamp Up is based upon the feedback we get now.

Yes. You get now from the people here. Yes.

Right. It's, it's a very, you know, it's that fin loop. It's a great loop of, of give us your feedback.

We create products, we iterate based on your feedback. We create products and it's, it's virtuous, right? Yes.

It, it really works. So, yeah. And what's really important about this kind of conference is that when you have a relationship with a customer, you talk about the, uh, issue that they have with the platform and what they are trying to do and all this.

But, uh, in here it's a lot more open. Yeah. It's really the, what they are trying to do in general.

What, what, what is their market and, and what are the technology challenges? And a lot of times they don't think that Jeff Frogger actually fit, but we find out that there is a lot of their, uh, issues and, and things we can help a lot In Their, in Their role. Well, it helps drive what you, what your roadmap is.

And there's no doubt. Now, there was several big announcements this year, uh, at the show, the Nvidia partnership, GitHub partnership, um, uh, uh, not endpoint. Security Runtime, Runtime security module for the platform.

Well, it's not a module. It's a PLA product. Yes.

Runtime security product for the mod, uh, for the platform. What we haven't really, it was mentioned, but it, it hasn't been as front and center, let's say, is a lot of the work going on with jfr ml, the MLOps, it comes out of a, a, uh, acquisition. Acquisition from Quack.

Yes. And we, if people look, if you go on tech drunk tv, we interviewed the CEO of, of quack when the acquisition went down. But Fred, talk to us about that.

What, what's happening in that arena? So yeah, MLOps is actually like, like I said, one of the changes that happened that makes our world and, uh, the tech world never boring. And I really, really map it to, uh, what happened in the early days with, um, mobile application.

Yeah. When, um, basically iPhone started and we said, okay, we can actually create application for the mobile world and, and provide application. Uh, company really didn't know.

So there was a bunch of either student coming and all kind of, because kids always love this and, and they love to provide. That's what they should have. Yes, Exactly.

And they ended up having application on the app store, on the play stores with the name of the company developed by a team that with zero control. They were compiling the mobile app on their laptop and pushing it Yeah. By themself.

And, and, and so suddenly they realized, wait a second. It's, it needs to have good, we need to have an application security And, and, and so on. Yeah.

Version, version follow on. Yeah. Uh, relationship with the customers and, and all this kind of stuff.

And this is exactly what's going on now with m lops, the data scientists, they used to do a lot of data science, a lot of providing analysis and, uh, and feedback to the internal of the company. And suddenly with the Gen AI and, and all this, they need to start to put what they are creating in front of the customers of the company and, and out there and really impact with the, I mean, a lot of impact on the outside and inside of the company of all the work that they are doing. And they never had any constraints of versioning, constraints of security, constraints of deployment, constraints of compliance and follow up and things like that.

And, and this is what we talked about, that more than 80% of the actual MLAI application that are trying to be developed, never make it to production because of that data scientists were never really trained or confronted to this problem of putting a reliable app application in production Discipline. Yes. And, uh, and with DevOps, I mean, DevOps is kind of a young term if you want it, but, uh, well, Patrick Dubar is here, he'll tell you 2009.

And, uh, but we learned a lot Yes. With, uh, with the DevOps practices, what is good, uh, quality gate and, and good, uh, monitoring and practices and agility on, on putting, so we can totally reapply everything that we learned about DevOps to the ML model. Now it's very different.

Even like mobile AppSec also a mobile app is very different than a website. Yeah. And, uh, and so you cannot have the exact same processes and tests and Code Everything and security tests.

And so we need to use the practice and the best practices, but redevelop the tools and, and the system and uh, and, and the implementation. So something I saw in the, in the keynote yesterday, I, I think it was when, show me it might have been when Y was up, was that really this, this, uh, data scientist represents a new personas for us to to, to help. Yes.

'cause we're here to help, right? Yes. And, uh, now, so it's not just making the app better, like in mobile AppSec, we, we, we needed some structure to mobile AppSec.

We needed to, it, it didn't have to mimic exactly, you know, web AppSec or something, but there needed to be a, a real structure to it. But you could have a, a developer who would do mobile development on this project or maybe on the next project or something Else. I'm Yes.

That is A data scientist is a different animal. Yes, totally. You're you're a data scientist.

You're a data scientist. Yes. You're always a data scientist.

And so, and their world is changing. Yeah. Yeah.

Yeah. Well, they, they need to be taught. So what, what it mean to do ops, like, by the way, not so long ago, developer, they used to write code and putting their code in production was not their problem.

Right. Throw it over. That was free DevOps.

That's the way data scientists behave until now. I Think So. So, so they need to say, wait, my, my model is actually going, what's really us, by the way, for Jfr is that, uh, even for mobile AppSec and web app and stuff like that, the developer, they deal with source code, they deal with files that they can read and, and all this stuff.

Data scientists, you never read the list of, uh, weight and biases of your, uh, no, I understand. Everything is binary. Everything is managed as a, as a tool that you read and, and you debug and, uh, so you, you don't read the code.

So everything is binary right from the get go, which is really good for, for J folks. So we can really provide this environment for the data scientists that mimic a little bit what git, uh, uh, provide for, uh, for developer, for provide developer and things like that. So, But that makes sense, right?

Because if they're gonna be part of the community, and that's really what it is, they're becoming part of this community, they should follow. If it works for them, it'll work for them too. Yeah.

Yeah. But it's, we still need to listen to the way they work and, and, and what is their problem and things like that. It's, it's quite different.

Mm-Hmm. But yeah, the best practices and, and the constraints. And usually once they start to see that they can have a lot of really good automation and validation and, and, uh, and the repetition of, uh, checking the quality of a model, for example, is something a little new for a lot of data scientists.

Yeah. You know, they used to do this experiment and look at them and say, ah, this one is the good one. But at the end, when you are doing A-C-I-C-D platform and a pipeline, you said you, I need to automate my, uh, it's the good one.

Otherwise it'll never get, it'll be left behind. Right. You, yeah.

Yeah. The knowledge of what makes you decide needs to be captured in code and needs to be captured in the automation. Understood.

And, and, and things like that. And usually as any kind of technologies, even the data centers or developer, once you start to see, wait, I can actually automate a lot of my, my need and my code, they are happy. They, they, they like, uh, they like this stuff.

And, um, so It's a beautiful thing. So for people out here, they're not here, how could they get more information on jfr ml? Yeah, so it's, uh, already a ga We did, uh, as always integration and, and it's, uh, so we integrated frog mail in, into the, it's now on the, on the cloud, on the platform, on our cloud platform.

And, uh, you can activate it and there is a lot of different modules. So you can start just with the exploration and, and, and management. And it goes all the way to monitoring and, uh, and management of to the Full tank.

To the full tank. com, check out jfr ml. Yes.

Look, this, first of all, welcome data scientists, right to the Jfr family. Yes. Second of all, we're here to help.

Seriously. And, and, uh, it'll be interesting to see how this Looking forward, You know, plays Out, looking forward. We see so many models out there, and I really believe by the way, that, uh, LLM feels like a single big, but the success of AI is in the multiplication of model and the different model, the model that verify the model and the output and, and all this kind.

So we need more well tuned and well trained model to kind of control and make really our, uh, experience of AI agent and, and all this AI system a lot better than what we used To. Agreed. Fred, thank you so much.

You're welcome. Fantastic event as always. We'll see you soon.

See you soon. Alright. Fred Simon, co-founder, chief data Scientist here at, uh, at Jfr.

We're here at Jfr Swamp Up. Our, our coverage of day two will continue in just a little bit. Thank.