Jussi Kiviniemi, founder and CEO of Hamina and a self-described Wi-Fi nerd, live-demos an AI tool that turns low-quality floor plans into 3D digital twins in seconds, with no site visit. Built for designing wireless networks, it detects scale, walls, doors, windows and materials, stacks floors into whole buildings and recognizes domain features such as warehouse shelves; Jussi invites the audience to suggest other uses.
“So we have no access to the site. This needs to happen in seconds. And all we have at our disposal is low-quality floor plans.”
“sure, it's not as detailed as a digital twin model using Gaussian splatting and stuff like that, but also it was created in five seconds, right?”
“It's not off the shelf, a large language model. It's our own AI science built specifically for this purpose by Finnish engineers”
Jussi Kiviniemi showed Hamina's tool turning a single floor plan image into a 3D model in seconds, detecting scale, walls, doors, windows and wall materials automatically.
According to Jussi, the model is trained on many kinds of environments but is not yet optimized for oil rigs.
Jussi said it depends on the environment; heights can be tweaked and sometimes need manual correction with built-in tools, with more automation planned.
Transcribed automatically (ElevenLabs Scribe) and lightly edited. It may contain small errors. Click a time to play the video from there.
Jussi Kiviniemi: Thank you so much. Thank you so much. Thank you guys for joining. Um, I mean, what could go wrong if you try to do a live demo in front of, uh, 400 people? Uh, we will, we will find out.
Jussi Kiviniemi: No. No, no sound on this one. Um, let's see if the, uh, we, we get the HDMI to work. There we go. And, uh, we are off to the races.
Jussi Kiviniemi: That's excellent. Uh, I'll still hook up my display, and we'll be good to go. Mirror. All right. So, how do we create 3D digital twins out of almost nothing, out of thin air?
Jussi Kiviniemi: Um, I'm Jussi, and I'm a, I'm a Wi-Fi nerd. Let's face it. And, uh, and then you're like, "Okay, what does a Wi-Fi nerd do in a conference like this?" This is, this is what we do on a daily basis, actually. This is what true Wi-Fi nerds look like.
Jussi Kiviniemi: Um, so, so why are we here? So we've created almost accidentally this, uh, 3D, uh, AI-assisted 3D modeling technology, okay? Uh, that we use for our Wi-Fi nerding stuff of us and our customers mostly. So, so thousands of people use this stuff in the Wi-Fi space, but nobody outside of the Wi-Fi space has e- even seen it. So let's take a look at that.
Jussi Kiviniemi: Uh, I gotta hand it over to Michal Gula. A few days ago, I had never heard of this conference. I had no idea this kind of stuff existed, a few days ago. And then I saw Michal, like, post something on LinkedIn because he has an enormous network. I saw his post and he, uh...
Jussi Kiviniemi: and I, I just messaged him like, "That's really cool." And M- Michal said, "Well, your stuff is cool too. You should come to 3DISE." And I was like, "That's in three days, man. You're crazy."
Jussi Kiviniemi: And then it just happened. So here we are. Uh, anyway, so, so a round of applause to this guy, Michal. Like, you're very fearless, aren't you? All right, uh, so here we are.
Jussi Kiviniemi: And the technology we built is a combination of AI models, uh, that creates 3D, uh, digital twins from almost nothing, all right? So for our, uh, use case of this technology is to design robust wireless networks. So you need to kind of predict the wireless signals as they go through shelves in a warehouse or walls in a, in a building or stuff like that. But this can be used probably for 1,000 other use cases. We just haven't figured it out.
Jussi Kiviniemi: We've been focused in our Wi-Fi nerding stuff all this time, okay? And the need for our Wi-Fi nerdy customers is to instantly, in seconds, create a d- uh, 3D digital twin. Uh, and they don't want to input any of this stuff in there. They don't wanna put in the metric scales, the building boundaries, the walls, wall materials, doors, elevators, staircases, stack the floors together on top of one another, but they need to have that 3D model. So, uh, and then they have a couple of constraints.
Jussi Kiviniemi: These are needy people. We're Wi-Fi nerds. We are, we are very needy. So we have no access to the site. This needs to happen in seconds.
Jussi Kiviniemi: And all we have at our disposal is low-quality floor plans. So, how do we do that? Okay. Uh, let's, let's drive a, um, a live demo and see what happens. So I was, uh, I was in London last week, and I just ma- went to Google and said, uh, "Find me a hotel, uh, floor plan from London."
Jussi Kiviniemi: And the first one, lo and behold, was Ritz, uh, floor plan from the Ritz. And it's the ugliest, uh, like hand-drawn floor plan from the 1800s. So, you know, uh, why not use that, right? So let's, let's, uh, save that image on our hard drive, and then I will open our application. So never mind the Wi-Fi nerding stuff.
Jussi Kiviniemi: That's almost the whole application. Just mind, uh, the 3D modeling stuff that we've created. So what I'm gonna do is I'm gonna drag that Ritz hotel floor plan in here. And what happens is AI starts to figure things out. You may not see this at the bottom, but AI already figured out the scale, metric scale of the drawing.
Jussi Kiviniemi: So it says, robot says, "This is approximately five meters distance." Then what else does the robot say? Robot says, "Okay, I have found this, uh, exterior extents of the building. Uh, do you want to keep these?" Yeah, they look pretty good.
Jussi Kiviniemi: Uh, AI also says, "I also found the walls of the drawing. Uh, do you want to keep these?" So it's, it just doesn't just find the walls. It fi- finds the windows, the doors, the different wall materials even. If they are drawn with a different pen or a different thickness, it automatically categorizes them as, as a different material, stuff like that.
Jussi Kiviniemi: And then the end result is indeed a, a 3D digital twin out of freaking hand-drawn picture from the 1800s, okay? Uh, so, so then you're like, "Okay, Jussi, but you're, you're probably just, you know, pulling my leg. Uh, y- you know, this can't be real. Give me something more tangible. You probably just, you know, mashed this all up."
Jussi Kiviniemi: So how, how about we go to CUBEX? You know where this is from? That freaking wall, okay? Uh, so, so, so I took pictures with my lousy iPhone of, of every fire escape plan I could find, and I also tried to go to floor number three, but I almost got arrested, okay? So, so, uh, uh, do not go, go further than floor number two.
Jussi Kiviniemi: All right. So, uh, and it doesn't stop at one floor plan. So I c- I can also punch in, uh, the auto construct building, and it takes a look at all the floor plans and then stacks them on top of one another. It orders them correctly and builds not just the floor, but the entire building. So again, let's see if we could make this out of a...
Jussi Kiviniemi: Yeah. Yeah, that's pretty good. So, so there's the windows, there's the doors, there's the bathrooms, uh, in, in there. Uh, that, uh, Mr. Ultrapetch is totally not checking in, uh, as, as we heard.
Jussi Kiviniemi: So, so it's all here, right? Uh- But then can we also see this in, uh, in its full 3D glory? Let's, let's see how our, uh, 3D mapping thing is doing. So, uh, let's look at this guy first. This, just this floor, and then all the floors.
Jussi Kiviniemi: So there's the whole, uh, all of what I could visit in CUBEX Centrum created in, like, three seconds. So sure, it's not as detailed as a digital twin model using Gaussian splatting and stuff like that, but also it was created in five seconds, right? Uh, and this is just getting better and better and better. What you're seeing now is the worst this will be. What you will see next year is a lot more richer, a lot more accurate, a lot more detailed solution should Michal not throw us out from the conference after, after this disaster.
Jussi Kiviniemi: So, so we'll see about that. Um, okay, uh, then let's go to, uh... L- let's do something else. Okay, sure, it can detect walls and stuff like that. Uh, but since we are in the Czech Republic, h- I, I went, uh, and searched for beverage warehouse floor plans.
Jussi Kiviniemi: So, so, uh, how about a beer warehouse? Luckily, somebody has put, uh, put beer warehouse floor plans, these low quality JPEG drawings, also out in the interwebs. Uh, so, so let's actually smash that in here. All right? So I, I will actually say this floor plan is of type warehouse instead of an office or whatever carpeted space, and I will drag that guy into the canvas.
Jussi Kiviniemi: So what's supposed to happen, uh, unless, uh, you know, our system is broken with beer, but, um, what's supposed to happen is not only should the system detect things like walls and doors and things like that, it should also detect that these are warehouse shelves and then raise them up pretty close to the ceiling and then, then, you know, make the ceilings higher and stuff like that. And then if this was a data center, uh, it, it would build out the data center racks and things of that nature. So there is some, like, domain specific intelligence als- also in there. Uh, but yeah, uh, that's the, that's kind of the idea. And sometimes AI gets it, sometimes it doesn't, and it's never per- Oh, yeah, but it's, it's pretty good.
Jussi Kiviniemi: So, so it figured out, uh, k- kind of the shelvings and, and things of that nature, uh, out, out of a JPEG, uh, plan, out, out of the interwebs. So this is kind of what we do beside our Wi-Fi nerding. And, uh, I, I also did... By the way, if you wanna come to our booth, I also did the, the hotel we are staying here in Prague and, and various different hospitals and things of that nature. You can also find a lot of good floor plans from the interwebs.
Jussi Kiviniemi: Uh, so in one minute you can create using low quality floor plans without a site visit, pretty cool, uh, digital twins. What can we use this for? I have no idea. Please come and help us out. So, uh, hit me up on LinkedIn.
Jussi Kiviniemi: Uh, thi- this takes you to, to my LinkedIn, so just snag a picture. Um, and, uh, yeah, would love to continue these discussions. Uh, Michal was fine of a gentleman enough to give us also a booth here, and he's gonna throw me out. Can you play the symphony music as well? Uh, just like in the Oscars.
Jussi Kiviniemi: Uh, so, so yeah, come see us at our booth and, uh, you can also try it instantly for free, e- especially if you have a Wi-Fi network to design or something like that. So go to hamina.com and try it out. But thank you so much. Uh, thank you for having me.
Jussi Kiviniemi: And, uh, yeah, look forward to going-
Host: Let's give it up for Jussi Kiviniemi.
Jussi Kiviniemi: Thank you so much.
Host: No, st- please, please stay with me. I'm not kicking you, I'm not kicking you out. Please stay with me.
Jussi Kiviniemi: That's very kind of you.
Host: We can keep the, the... We can keep the thing on the stage, please, for a, for a, for a bit. We've got questions for you as well. A- actually, quite, quite a lot of them. Uh, so can we bring them up on the screen once again?
Host: And, uh, let's start with the top voted one. So is this tool working also for industrial sites, floor plans like oil rigs, vessels, or r- refineries?
Jussi Kiviniemi: We do have, uh, totally not my personal super yacht, for example, modeled as well. So, so feel free to come in. It's being trained on all kinds of different environments. You, you can figure, uh, s- so... Uh, but it has not been optimized yet, for example, for oil rigs, but come next year, it probably will.
Host: Very cool. We've got a personal question.
Jussi Kiviniemi: Oh, gosh.
Host: Are you okay with a personal question?
Jussi Kiviniemi: Totes. Totes.
Host: Let's do this. So how did you become a Wi-Fi nerd? What's the story?
Jussi Kiviniemi: I don't know. I stumbled into a cord, unplugged it, and then, then it just worked wirelessly.
Host: Okay. Fair enough. Uh, you were talking about AI. What kind of AI exactly?
Jussi Kiviniemi: That's a good question. So, so everybody touts AI, right? And usually when you find an AI startup, it's, it's like y- you put a crappy piece of software together with a large language model, and that's your AI startup. This is not that. It's not off the shelf, a large language model.
Jussi Kiviniemi: It's our own AI science built specifically for this purpose by Finnish engineers, that you can find a couple, a couple in here as well, and using our own data. We actually employ even our... the most expensive data annotation team in the world that sits in downtown Helsinki, drinks Don Julio, and, and just annotates the da- data away, so.
Host: What is Don Julo- Jul- Julio?
Jussi Kiviniemi: Don Julio is an expensive tequila that we serve to our annotation team, so that they keep up with the-
Host: Okay. Okay. Fair enough. Fair enough. Uh, last question.
Host: Um, how are wall heights determined if this information is not included in a drawing?
Jussi Kiviniemi: Excellent question. So it kinda, uh, it, it kinda depends on the environment. So we can do a little bit of tweaking on the wall heights. Uh, but come, come to our booth and you can see sometimes they need to be manually also corrected. So we have excellent built-in tools for that as well.
Jussi Kiviniemi: And next year it will be more automated as well.
Host: Amazing. Jussi, thank you so much.
Jussi Kiviniemi: Thank you so much.
Host: Excellent demo. Everything went just according to the plan.
Jussi Kiviniemi: Yeah, 100%. Thank you, and my apologies. See you later.
Host: Jussi Kiviniemi.

Founder and CEO of Hamina.
AI-Driven Wireless Tools helping network engineers perform better. Former Juniper Networks technologist and leader of Ekahau’s Wi-Fi Design Tools, he combines deep networking expertise with product vision, marketing instinct, customer focus, and a passion for strong company culture.