John Rodrigues introduces JEV, a decision-layer model that orchestrates multiple LLMs for generative UI creation. At Human AI Studio, it boosted retrieval accuracy from 72% to 95% and made UI generation near-instantaneous. The model was created by Diego Mera, co-creator of ChatGPT, under Tagspace AI.
Chapters
So welcome everybody. So in this one, I'm gonna talk about Jev. The Twitter is kind of blowing up based on this and it's pretty interesting and cool. So what I did today is I kind of took this, the Jev model that recently released and applied that to the generative UI and especially like how it's applied to like design systems. And then it just, this model is incredible. It's like changing a lot of things like reducing the cost and all that stuff. And everybody on Twitter are making experiments and like sharing. So I thought to kind of make this video today to break down what JEV is and how it's applied. And some of the research we did at Human AI Studio, I'm gonna like share that here to see how JEV can be useful. So basically it's a model, but it's not like another LLM that came out you can chat with. So that's why there's a lot of like confusion around like what this is, what it does. So JEV is basically a decision layer that exists between different LLMs while generating and things. So I caught, it caught my eye when I saw Chris from the Vercel Labs, you know, you know, you know, made this video and got like 1.2 million views. So here you can see by default, the generative UI is like taking time, but Jav is like quickly instantly making that UI. It's not just, it's just fast. It's just like making that decisions very quickly behind the scenes on like what model. And when you're building, like you have a lot of like you answer a lot of questions, right? Like you ask a lot of what model to use, what to pull, which component to pull and all that stuff. So Jav kind of makes those decisions like very fast. This is like very fast model. So here it's pulling the things from the design systems and it's like making those UI very quickly. Like even like, even in the blink of an eye, it's like making that. So I took this and applied it to our AI-ready design system, Florence at Human AI Studio. And I'll include the link to that as well here. And what before the retrieval was like the agent's ability to understand was like, like retrieve list was 72%. Now it went to like 95% by using Jev. And then it generated this dashboard like so quickly, instantly with the AI-ready design systems. And then AI-ready design systems are like so important nowadays because the agents or any LLMs or anything won't be able to generate without a proper design system. So after using Jev, it's like efficiency increased so drastically. So it was founded by Diego Mera. And who is the co— who also co-created ChatGPT. And then before it was like OpenAI, and he started this company called Tagspace AI. And they launched like recently, like a couple of weeks back, as depends on when you see this video, it was like a couple of weeks back. And so Jev is basically that decision layer that exists. So user asks something and instead of like depending on the users to, you know, make this decision, Jev makes this decision very fast. And then it also reduces the cost because it picks which model and everything is almost like orchestrated. The models. This is very interesting because how it's applied to the field of, you know, building products is like the user interfaces. Generative user interfaces is like ongoing research in this space. So that kind of makes it more possible now. What that means is like, let me go to that in a bit, but let me cover this one. So basically like if you are asking, so this model, what is this? It's a classifier model. It classifies. So if you're asking in the generative UI, so if you're asking When the, let's say a UI needs to be generated. So it also makes a decision when this UI should come, when to show refund policies, when not to show, which interfaces comes in this first moment. So it makes those design decisions very quickly as well. And then it picks the right thing for you. So usually user manually had to like ask all these questions and kind of generate this UI and all this stuff. So right now it generates UI, it makes that decision very faster. So, but it, It is, it sounds good, but there are also a few limitations to it. And like, how would you even know the user intent? And I'm applying this only to like the generative UI and design systems, but this can be applied to like building agents, automations, workflows, and like automating a lot of things. So there's, it's, and also like building like applications and like selecting the models and reducing the cost. So tons of applications to it. I'm speaking it more from the lens of like the product design and speaking from the more lens of like the building the products and things. So usually the LLMs, they generate, right? And then they're like pretty random, hallucinate and all that stuff. Javice is a classifier model. So it makes those decisions what needs to go where in order to like give you the better output and stuff. So it kind of decides like, does this need human review? Which UI, which is safe, which is a model, which model should you use? Is it simple? Is it complex? So it kind of makes this micro-level decisions very faster. So it's like owning the decision layer behind those models and things. So the outcome Some of this is it makes it faster and then cheaper. This is a huge— the cost reduction is huge. I've seen people categorize an email in like few minutes, thousands of emails and stuff. The automation becomes insane with this. And it's also more predictable and also accountable because it has that confidence score. So yeah, but this brings this question of like, okay, then if If agents are making the decisions about the user interfaces, then who's going to inform or give that context around what those components and components and the system, the design system is. So this is where the AI-ready design systems comes into picture. And at Human AI Studio, we did a ton of research on AI-ready design systems. We've been doing like, we also launched our course. So this is a very important topic and there's something that we are betting on at Human AI Studio because any of these models need that context layer. To, in order to like make those generate UI, to generate the user interfaces on brand without shipping AI slop. So, and this is also here, also Chris mentions about like it makes all those decisions based on your design system and you are like baking in all the designs in this design system. And if you want to like learn more about it, we also have, like, let me show you here. We also have a dedicated course on this. So it's, you know, self-paced course. You can go and take and watch. We also have like a community and then yeah, we also have like the community. And then self-paced structured course and more of this new things happen. We are updating. So I'm going to go in those as well, how to apply JEV in the AI-ready design system and things. So JEV is becoming like that decision layer, which is very interesting. And I started seeing the results like so quickly. So I went to like 95%. That's insane. So yeah, in the generative user interfaces, it needs to make the decision what UI needs to come when. Is it practical? I think it's definitely been heavily researched. Strong right now because yeah, you can generate UI, but knowing when it comes like even Jav models, models like Jav is kind of like solving that problem, but then still understanding the user's intent and things that still remains harder. But yeah, but this is making like, you know, it's much faster. It's making building products and generating the UI on brand without the drift is making faster. So it's very important to kind of learn about this AI-ready design system because that's going to be the core infrastructure to how agents are going to build. So design systems like gives you that infrastructure for these agents and then Jav makes the decision layer and then definitely there's human in a loop also directing and guiding this. So yeah, this is worth looking into. Jav is a model that's going viral on Twitter. Hopefully eventually will be other people will know and use and all that stuff. So yeah, Jav is insanely good and it's worth looking into. And yeah, if you want to learn how to build AI-ready design system, we have launched our course. You can go to humanaistudio.io/course where you can go watch where you can join the community and go through a structured learning program. And as the new update comes, we're going to keep dropping this tutorial. So the next, I'm going to make a tutorial on how to use Jamf in your AI-ready design systems and more. So thank you for watching. So make sure you are subscribed if you're watching on YouTube or if you're watching on— if you're watching this on Substack, so you get continuous update from us, from Human AI Studio.
