Pointr’s Post

Transcript

Happy Friday. I don't usually do this so wish me luck, but what I wanted to do today was we get so many questions on AI mapping, what it is, why have you invented it and what purpose it serves. I felt well the easiest way to explain it could be to just a quick video show the AI mapping product we have what we call it map scale inaction and then talk about what it does. So. But before I dive into that, I want to talk about maps in general. If I were to do this video, if I were doing this video 15 years back. A lot of things would be different, but and if I were talking about city maps and outdoor maps, the chances are the use cases or the value wouldn't have been very clear because you could drive home, fine. You could order food, fine. You can hail a cab, fine. And now Fast forward to today. With that outdoor maps infrastructure in place, a lot of those use cases, a lot of those journeys have transformed into much more digital journeys. You know, you order your food, you order your cab, you meet with up with people or you know navigate all using additional infrastructure and some sort of an application today. And I bet those watching this video will have used some sort of outdoor maps or GPS, you know, before you watch this video this morning or this afternoon, so. You know, the transformation has happened in the city, in the outdoor world. We believe, in fact, we're convinced the next frontier is indoors. Now if I'm comparing outdoor maps to indoor maps. The advantage outdoor maps had was you could literally take a photo from the sky using satellite effectiveness for satellite imagery is. And I don't want to underestimate the challenges. Still, it's, you know, you start with the photo has a lot of technical challenges that needed to be overcome to produce the maps. You're using something like Google Maps today. That's the advantage they had. And so of course, the challenges were still there. Now we don't have that advantage in the indoor space. You could try taking a photo, which would be a bit spooky. That just wouldn't work. It's not technically viable. That that's a disadvantage for indoor world, but the advantage that we have is most buildings in the modern world have been built by an architect. Have been designed by an architect using a password called the floorplan file, and those are typically your AutoCAD files. You know, some kind of a CAD format, DWG, DXF, so on and so forth. So we have a good starting point. We have those floor plans access for almost every building out there. But there's there's a ton of them and. Just to show you the the the kind of scale of the refresh your memory the if you will. This is the number of buildings. This is a nice view from our Boston office. Unfortunately not the best weather today, but. You know there are 1,000,000 if like hundreds of millions of buildings and that leads to that constitutes billions of floors. You're looking at billions of floor plan files out there. Now you may be thinking well this is if you have the data, surely it should be easy to transport. The answer is not every floor plan file is pretty much different. Even within the same companies real estate we see a different office managers have used a different architecture firm and will have come up with a different style and then say style. I just don't mean the office style. I mean the way the data is represented from the way you represent doors so hallways to entrances and exits and windows. Some have a lot of layers, some have some don't use layers, some use. Little scribbles on the maps on have legends, some have scales, some don't. So it's really you're dealing with very different files and you multiply that by billions of floors out there. It becomes a very difficult problem for a human soul or for humans to Marian soul. And that's exactly what we're tackling with AI mapping our tool called Map Scale. So. Before I, I, I, you know, as I mentioned at the beginning of video, I want to show you the product. So I'm gonna talk less and then actually show your product. So I've launched sales demos, which is our adjusted demo environment for our, what we call Poetry Cloud. So let me just log in real quick. This is the demo Gods are with me. This is our back end system, our content management system. I'm just going to create a quick site very quickly. It's called the Friday site. And then let me just pick. I don't know if that's pickets. It's actually really close to our office. They just picked it, assumed this is our corporate campus. Right. I just created a site. Let me go ahead and create the building. Devil buildings. Again, just do some rudimentary. Right, so let me go into this building. Now I'm just going to create a level. So I'm going to say this is my office. So I'm going to say I'm about to upload office 47 level one. It's called level one, and then let's call it L1 for short. So as you can see, you can drag and drop the DWG DXF for a PDF file straight from the dashboard. We also do Geojson but. Most people, most folks don't have Geojson, so we in the most. More often than not we can have DW, DWG, DXF, PDF being imported here. So let me just go ahead and upload this file right here. So what you're going to see first? Is going to see the roll version of the file and hopefully that will help you explain, help you understand. So this is what the floor plan is actually a cleaner floor plan looks like. Has a bunch of details, but. The trick is you don't want all of this to be shown in your map. Probably you have a lot of data that you don't want to show, and as you can see, they all look the same. So it doesn't know which one is a room, which one is a hallway. It just all looks like a bunch of lines and then squares and circles. Mapping is converting that into something that's. A human can easily understand, just like Google Maps have done, the cities. You don't see different types of roles except you just see them as just one Gray line and you know, those enrolled in the map. So our goal is to now convert this into understand what? So couple of stages really. First clean up the map. The skeptics of scribbles get rid of everything else around it should often noise basically. Or again, the floor can may have water pipes that you don't probably don't wanna show to some users. So you want to get rid of the artifacts and the details that you don't want. Secondly, to classify, so this is a table, is the chair, is the roof, and you'll be surprised we have an editor of this actually. If you haven't seen it, I would highly recommend it. But talks about challenges of even something as simple as a door. Look so differently across different floor plans. So now it's created that we just go in. So when I go, it's gonna say hey, our maps is gonna kick in and say this is eligible for map scale. Show you to try. So of course I'm going to say let's go. For this demo, so it's gonna take a few minutes. Will probably Fast forward this in the editing, but. Effectively you. Cleaned up the file you then classified and then the last piece is the metadata. So for an office, knowing meeting room names or the disk names can be very important. Or for a hospital, you know different room names, different office numbers, or for airports, different gate numbers, of course. So being able to get that data out of the floor benefit and most floor plans will contain some kind of metadata. Excepting those and associating it back with the map is is super crucial, so that's what our AI also will do. And the way we train our AI. We didn't come up with this because Chachi PD came out and we had to have an AI solution. We've been working on this problem for close to a decade actually. And we felt, you know, we can do a few buildings here and there with our mapping partners at the time. But the second you go beyond say 1500 buildings, it just doesn't scale and you throw up, you know, throwing more humans that this is not the right solution. So you felt can be used software for this. And then we effectively created a deep learning engine. That learned how to mimic a human league. And learning how to learn. Is very difficult. I don't wanna take the credit was our our entire team has done a fantastic job in close collaboration with our delivery and design teams and of course our engineering teams. But they managed to create this engine. And the beautiful thing about engines, it only gets better the more we they, the more data we feed. After all of this pitch, I hope that the result will come out nice. And you may be thinking. Surely this is a a sample file that you know the answer is no. Actually this could be any file. And that's the power of map scale for us. We, in fact we will be launching this on the website very soon. We're aiming for later this month or early September where you will be able to just drag and joke around fires to see the result for yourself. And you're very excited about the feedback you're going to be receiving. It's our model has come a long way in the last five years especially, and we now hit around 90 to 95% accuracy. Just in a few minutes. And you know, being pulse integrating, mission level up, sorry, mission learning DevOps together, learn from the additional fine tuning as well. So it's only going to get better and get better at the faster pace. So that's what AI mapping really means for us. So it's going to take a few more minutes. So we'll see the result and I can show more of that, more on that shortly. Right. So it took 7 minutes as it says here and it did what I was. I hope it did what I was suggesting it would, which is. Clean up the map, let me zoom in. Clean up the map, classify everything from hey this is ever extension, this is some, this is a restroom. This is a workspace and just a meeting room and that's why they have different colors and then also extract the metadata. So I'll hit continue. And this is our demo environment. This is actually slower than usual about, you know, 7 minutes is still not bad. And all of this is available via API, so most of our partners don't use the dashboards, they use just, they just feed their data through the API. She'll talk about in a minute as well, but once this is once I confirm the result, you'll be able to dive in a bit deeper. So what I'm what I'm able to do is I can actually toggle the original floor plan to see how Valve. It's captured the floor kind so I can, you know, compare them twice for myself for my reference. And we're actually working on a version where you're able to immediately point out bits where it's not sure about so that if he needs to review, it can show you. But as you'll see, it had these chairs in the origin of floorplan desks, for example. It had this fantastic example. Actually. You see, there is a, you know, to the human eye, this could just be a window where a door. But actually in this case, these were the cabinets and our AI was smart enough to actually ignore. So if you just did pattern recognition and thought that was a window, well you would have a tough time trying to understand why you had a window next to the stair, a bunch of disco of Bank of tasks. But in this case I was smart enough to not do that and not fall for it. Be and you can also see for each desk. So if I just click on this for example, say able to get the ID. You're also able to eat what additional date details that we had in the desert at the ID and his something. There's additional name for that specific task if I click on something or if I go to this maybe meeting room. Right here. You can see it's actually labeled as a meeting space. And if again, if I click on the metadata, it has an ID, is that a name? So it chose to show the name and we're doing some interesting work with. We haven't released that. We will announce next month in our quarterly September release. Talking about how you're going to use LM to even improve this but. At a very, very quick kind of glance, this is what our AI mapping can do. So what would have taken a human at least a day or two to produce this map or I was able to lose in 7 minutes and. On that note, and of course I can also add, you know, if you want to do wayfinding absolutely no problem, you can immediately auto generate old routes. And the building. Safe and ready to go. So that's how quick we can digitize the building. In fact, we have some parts we used to tell us it would take them two to three months to go from a normal building to a digital building. With this, you want to lower that down to just minutes. That's what they're supposed to be behind how we can do the world's largest deployments in the space today. And I'm really excited about what else are you going to the house in the in the coming months. But just to give you about some idea about the scale I will have to. And maybe some of this is if I just go here, I say ohh. Will have to invest unless of tests and it just say you know what, show me what napkin has done since the beginning of this year. And then today is the 9th. If I just do since the beginning of the year. Of how much? We have captured, you can just say, a quick world map of all the buildings, all the floors that have been digitized using AI alone. And we're only getting started. So this is the city gives us confidence that we can tackle that billions of floors problem in a different way than the other scenario and tried. And we are very excited about the advertising their partnerships of you're going to be announcing more and more of that good stuff in the coming months. At the moment we are able to do 5 to 10 million square feet a week. In fact, we can do maybe more, but that's kind of what we are currently doing with our partners. We're looking to just add make that you know, access that even further so. Ethnic lens I just want to show you what are what you mean by AI mapping AI is easier said than done these days, but I felt a quick demo of it would be helpful. So hope you liked the video and hope you enjoy the rest of your fine. So thank you for listening and if you're in the feedback, please let us know drop in drop a comment. I again, I don't usually do this so I don't have a subscribe button or something else, but you know would love to hear better like this video and then what you what you you know what feedback you have for a mapping and. As I brought up, very soon this will be available straight on the website so you can drag and drop any floor plan and then see it for yourself. So until then, stay tuned and enjoy the rest of your Friday. Thank you.