Projects. Rework is, every time, but, I think it's 1 in 50. There's, like, water problems. Like, a pipe is misinstalled. Yeah. And that's, like, tens of 1,000 of dollars of cost. Yeah. Right? And, like, you could've just detected that at the install stage. So that'd be the ultimate goal. It's like some device can detect when there's, like, a yellow flag. Obviously, they send it to the cloud and have to further review something like major. But I think that's, like, our, like, our path. We're avoiding behavior change to start, and that's just, like, we're just, like, punting on the pain. Like, getting people to wear devices is actually a natural thing. Actually, this this really well with so one of our neighbor in Hillsborough, Lily, her family is very big in commercial and private real estate in Bay Area. They own a bunch of house in Hillsborough and those commercial real estate. Like, her, like, her parents just, like, do a lot of buildings. And they also have a manufacturer set to to do, like, the construction, like, material bunch of things. So, she told me, like, this is her like, the moment when she saw this device without me saying any any single word, she she saw my words to her party. And she's like, wow, this will be so useful for my family business. She's like, we we want all we want the construction worker to bring this to document their workflow so we know what is going what is right, what is wrong. And I I didn't even need to explain, oh, there's a camera, there's a voice voice or something. Like, I just have I I really want them wearing this, and we can use this data to train new workers really well. Yeah. Yeah. And that that's exactly the states that we looked here for. You you just want the video off the scene or you want to do other shit? I mean, right now, we we're we're doing audio and visual audio. Yeah. Audio and video. Yeah. We we support, like, there's audio, there's a video, there's voice input, there they can have a talk with you. Mhmm. Yeah. Yeah. The training is interesting too. I mean, feel free you guys can go find out some of the AI pieces, but, I think the medium term plan is to edit the work in real time. Mhmm. So if you do identify something that is, like, the angle is not right or the the the the pieces, you know, misinstalled, like, to inform the worker in real time that that is is where we wanna go. Yeah. Like, that ultimately would save, you know, 1,000,000,000 of dollars of of waste. So the device could talk to you. You'd be able to double check that. Yeah. Yeah. Yeah. Ideally, there's visuals in that as well. That's why you wanted the screen. Yeah. The screen, ideally, would be nice. I mean, we're still, like, like Is that why the in state is probably AR glasses? Probably. Yeah. Exactly. Yeah. Yeah. I mean, it's been, like, who you know, like, the time horizon on that's long. Like, we're not sure. You know, even just even just notifying the manager that something's off is is actually worth a lot. Could you, you could still prompt the user via audio, like, hey, double check that. Yeah. Pull up your phone and the visual could be a phone easy thing to Yeah. Even for now, like, so what we're doing right now is like progress updates. But there are certain progress updates that are more important than others. So even now, something that says, hey, you missed the I know HVAC is a key thing. You didn't take a picture of the HVAC, like, going to the HVAC. Right? So, yep. So non visual is also quite useful as well. And the other thing is, I think about how to build a model in the device, like, describe the router that the app on the phone. Yeah. Yeah. So, basically, like, there's multiple parts to our back end. Once we have a trained router, depending on your, your scene and your question and your your specific use case. For example, it's like, oh, this component in the right size and, like, like, a query our internal database to find a data sheet for this. And it can it routes different, like, back end. Some of those are, like, more, like, reverse search. Some of those more, like, courier than this person before. So we have a trained router based on first person view and video data, and also image sequence. So besides that, we also have, like, a fall agent workflow, that Raj Rajvi building on, which using a basic graphic structure to search for all index data, including location, people, conversations, audios, time, all of those, and events. And along with that, we're also training, like, a visual model, that is like a like, that is trained down, like, action sequence based on the real world, and is able to understand if a person come into the room, leave the room, and capture those skills. And, we also have, like, earlier data, we build this little on on device, smaller model that able to do privacy filter and facial detections. So that is to avoiding, like, in case this person wearing this to the restroom, it stopped recording. So it knows that the some data should be destroyed on-site, like, on the device, not even uploading. And also tracking the motion, we have IMU on that. So it knows if this person is moving or if the it detect key frames of, if a person coming, if the person left, like, differentiated enough from the past frames. In that way, we know that the data should be, like, upload or take a keyframe. So Got it. Yeah. What do you think is the the the most likely first use case for you guys? Yeah. Yeah. So basically, like, we're right now, like, is focused on, like, b to b to c sort of, like, consumer angle a little bit. We're we're, like, so we're testing with a lot of initial, like, users. The 3 largest area we see have, like, overlaps. One is, like, actually, like, just fuel sales, in person sales. Another one is and entrepreneurs, for, like, self improvement networking. And last one is, like, conference, and event goers. And there's also a huge amount of, like, people reach out to us regarding, like, actually real estate, like like the 3, initially, like, the 3 inbounding demand to us is all actually in in around real estate. Exactly because they have a lot of visual a lot of visual information. Actually back in college, was over over 20 years ago, my one of my first research I did in college was AR with real estate with a professor cross between CS and civil engineering. So we're actually, like, using, like, a lot of time, like, Bing server, like, Bing to measure, like, other infrastructure, like, a comp I write CV algorithm to measure if the building was, like, moving a little bit and, like, to to do that angle of how long, like, the steel like, the steel pipe is, things like that. Yeah. Yeah. And then on a conference and you go to CES, go to head, and you take my own advice.