Spring Boot

So he went off and worked on the problem.

And then the Spring Boot people came and said, you know what? We would really like a J-SIG command.

And somebody else said, or maybe an MX Bean.

And so Matt said, okay, I got that.

You can take a checkpoint, crank another training run on top of it, and then take the resulting product checkpoint out the other end.

So that's called the linear workflow.

And that may be useful in a certain sense because if the customer doesn't know in advance where to stop, if the customer is willing to start with the first checkpoint, which might be just the spring boot quick start and nothing else, no other business logic or anything, That might be useful for many cases, but then the customer gets an appetite for the startup performance and says, okay, now I want to add my business logic in.

So the customer will run a combined linear workflow training run with first the Spring Boot stuff and then their own business logic and then Check Point after that.

So he went off and worked on the problem.

And then the Spring Boot people came and said, you know what? We would really like a J-SIG command.

And somebody else said, or maybe an MX Bean.

And so Matt said, okay, I got that.

You can take a checkpoint, crank another training run on top of it, and then take the resulting product checkpoint out the other end.

So that's called the linear workflow.

And that may be useful in a certain sense because if the customer doesn't know in advance where to stop, if the customer is willing to start with the first checkpoint, which might be just the spring boot quick start and nothing else, no other business logic or anything, That might be useful for many cases, but then the customer gets an appetite for the startup performance and says, okay, now I want to add my business logic in.

So the customer will run a combined linear workflow training run with first the Spring Boot stuff and then their own business logic and then Check Point after that.

We don't have that luxury now because as we look at this AI-assisted exploits, and we talk about chaining exploits together, chaining exploits isn't anything that's new, but AI has really accelerated the pace at which Those things can be discovered, but then also can be utilized to actually assist in the exploit itself.

And so when we think about these Spring applications and we think about how we've got obviously vast numbers, one of the most widely and most used frameworks around the globe, and Spring Boot clearly being front and center for us as VMware and Broadcom.

How do we begin to really think about how we respond to this and move more towards that proactive patching? I'm patching ASAP the moment it's released and getting almost into that continuous model for security versus the old school triage and we'll figure it out when we get there kind of approach.

It's great.

But on top of that, the position, the way that those CDEs, the way that patches are released, they are released before the CDEs.

The patch is always going to be there depending on how fast you're looking for new versions.

We've got customers that have had the patched version of Spring Boot or some other Spring Project on its way to production before the CV gets released or before the announcement of the new version.

Where customers are like, hey, Deshaun, I got a new version of Spring Boot 4.0.5. What's going on? I don't see an announcement for it.

So the idea is if you have a model and you can package it using this format or obtain it getting this format, I just downloaded one from Hugging Face.

You can run it right inside the JVM, so inside your Spring Boot application.

So you get really low latency for in-memory inference that doesn't require additional data.

And if your inference does need extra data, like let's say you're getting a request and it's a username, so you need to pick up the profile.

And yeah, you've shown a clear path for how companies that are running these types of MLOps types of models can now run it faster, closer to where they need

to be.

Yeah, there was another angle to it actually, because there is the really low latency inference, but then there's the idea of like, how do you get your model? If you're going to put it in memory in a bunch of Spring Boot applications, how do you get it there? And then if you have a model update, how do you get the update there? And Gemfire actually worked perfectly for that as well as a model store that can store data in any kind of format, has update events, and then you have spring listeners, and the whole thing just kind of like magically works very beautifully, not only for this case, but for any kind of artifact distribution.

So it was kind of like a two-for-one.

And, you know, you can kind of do spec driven development where you can tell claw to say, Hey, check out this repo, like, like write me a spec of this.

Like if I were in template medicine, spring and spring boot or whatever.

And also, if you wanted to implement this in OpenClaw itself, there's so many things.

It's not enterprise friendly.

But it's just fun things to show what you can run on CF.

Well, and here's something

that I think people are missing on this one, right, is specializing these bots because you brought up, you know, hey, how do you write this in Spring Boot? Let's go back to conversations that we've had with Deshaun and Dan, right? Deshaun's a big, big fan.

You always keep your Spring stuff up to date, and there's a lot of tools.