Summary

---

Summary

An introductory meeting between Flavius Pop (founder of Abonito.ai) and Joe Bartels (founder/CEO of Phlux Technologies), connected through a mutual contact named Peter. Both are founders of early-stage hardware/sensing companies with overlapping interests in industrial monitoring and edge computing. No specific deal or collaboration was agreed upon, but they exchanged ideas and left the door open for future conversation.

Key points

• Flavius is building Abonito.ai: small IoT devices for industrial machines that capture vibration data (predictive maintenance), spindle utilization, and audio notes from operators — all surfaced through a dashboard and chatbot interface.
• Flavius is also a research professor working on PMUT (piezoelectric micromachined ultrasonic transducer) sensors, exploring compact touch/pressure sensing for robot grippers as a potential research application.
• Joe is building safety sensors for robots — vision-based (color + 3D cameras) that eliminate the need for robot fencing, with an analytics layer tracking robot uptime, downtime causes, and worker behavior near cells.
• Both companies deal with the same edge vs. cloud data volume problem — Flavius generating ~30 GB/machine/month of vibration data; Joe dealing with high-bandwidth image streams.
• Flavius is targeting on-premises deployment for sensitive customers (e.g., semiconductor fabs); uses a gateway architecture to route data either locally or to cloud.
• Joe noted that his larger customers typically require on-prem; smaller ones are fine with cloud.
• Joe mentioned Picnic Robotics (Colorado) as a possible contact but wasn't sure of a warm intro.
• Both discussed heavy use of Claude (AI coding assistant) for productivity. Joe described using multi-agent teams in Claude (via VS Code) for tasks like requirements management, with specialized agents for safety, mechanical, electrical, FPGA, and standards review perspectives.

Decisions

• None.

Action items

• Joe - Send Abonito.ai website link (already shared: abonito.ai) and keep Flavius in mind if relevant contacts come up.
• Flavius - Look into Claude multi-agent team setup based on Joe's description.

Follow-ups / open questions

• Would there be any synergy between Phlux's safety sensor data and Abonito's operator knowledge capture? Left unexplored.
• Joe had vague ideas floating about possible collaboration/referrals — nothing concrete surfaced.
• Flavius's PMUT gripper sensing research: Joe couldn't point to specific robotics contacts currently using or needing such sensors.
• Possible follow-up catchup in ~1-2 months suggested but not scheduled.

Transcript

Hi, Flavius. I'm having problems with my camera.

Hello, hi, how are you? Hi.

I'm having problems with my camera. I... It was working five minutes ago. Now this is unavailable.

So.

Oh, no worries at all. Where are you based?

I'm in Pittsburgh.

Oh, Pittsburgh. Okay, cool. Yeah, I used to live there almost one year in 2016, the whole year. Yeah.

Oh yeah, I was here then. Yeah? Okay.

So now, so you are in the robotics department, is that right?

Yes, I did a PhD at CMU and graduated in 2019. Yeah. And then, uh, Did a post-op there for a couple years while we started up the company. And now I'm here running a robotics safety company.

Nice. So is that Flux? Yeah. Okay. Yeah. So Peter, he wanted to introduce us because I also started Flux. a company a few almost a year ago at this point, but yeah, not in robotics definitely. I can tell you a little bit about what you're doing, but I'd like to hear also more about what you guys are doing. Um... Sure.

Yeah. Why don't you start and then I'll...

Yeah. Yeah. So what we are doing is we are creating small devices like little boxes that you can attach to industrial machines. For example, one concrete example is a machine shop. They have a lot of CNC machines I placed those boxes on top of the CNC machines to capture a few things. One is the classic vibration to try to run some algorithms to do preventative maintenance. So sorry, predictive maintenance.

like look at harmonics and things like that. But the main thing actually we noticed people are interested in that domain is just spindle time, like how often the machine is being run. to give statistics on the factory utilization Because a lot of the time these machines are idle and they don't have an easy way to understand how like what's the percentage of idle versus actually running. And the other thing, we added a microphone to allow operators of those machines to take notes like an audio note, like you know those transcription services that you find online now.

So basically we do the same thing but for factories so that we can capture knowledge from operators and the type of knowledge we are trying to capture is like if they notice something that goes wrong with the machine, they notice weird sound, they will say it to our device so they'll take a note instead of writing it down. Right now they have some softwares depending on what type of factory it is but they always have to log in and type it so it creates some friction and the operators usually don't like doing that.

So a lot of it's missed. Um, For example, also notes about maybe they need to leave a message for the guy for the next shift. That's something that is also... sometimes missing So then we classify those nodes, right? So we put them in different buckets. So then you get the dashboard and you have everything about your machine that you should be paying attention to. And then if you find some, if you had an issue, you solved it and then in the future you want to trace it back, there is an easy way.

We have like a chat bot on the platform where you can just ask a question and then go find out the past issues or past notes without you having to go sort through and look in a spreadsheet. Um... And ideally, and we have, we just started this like a few months ago, this application actually, um, We are trying to correlate the sensor data, the vibration, to the notes from the operator. For example, if the operator noticed some weird sound at some point, Was there something in the signal?

before they notice that. So that's something, a different approach to predictive maintenance that people are not doing. People are mostly looking at signals only. sensor data, not really putting the operator in the loop. So we're trying to see if we can Uh, noo... Get the different side from that Interesting.

Cool. Cool.

Yeah.

Yeah, there's a lot of...

Um, interest in that predictive maintenance and how to minimize downtime and Or plan the downtime because you know it's going to go down. Minimize unplanned downtime.

Yeah, exactly. Yeah, cool.

Yeah, so now, so yeah, besides that, I'm also interested always in different applications, different type of factors or use cases. For example, now I'm talking with, a small semiconductor manufacturer here in the Bay Area where they use robots to move wafers from one chamber to another or from one tool to another and they are interested in putting a couple of devices on the machines themselves to check vibrations plus a few in the robot arm.

They haven't told me a lot yet. I just talked to them like a few days ago. But it seems there is some interest there. Probably if they want to move forward, they will tell me more. You know, they're super secretive. Yeah. So... Yeah, I don't know exactly what the application is, but they want to put it on a robot arm. So let's see what they come up with. Interesting. Yeah, and for them especially, it's really needed that everything runs on the edge.

For them, they're super secretive, right? So they don't want data to flow out. So they want everything to run on the edge. So now we are building our platform to run both on the cloud hosted by us and for like smaller customers like machine shops that they're not so sensitive and then full edge deployment for for semiconductor fabs, for whoever is going to have that type of need.

Yeah. So what do you mean exactly on the edge? Because when we talk to customers, I'm not sure they know what it is half the time.

To them it's like, or to me really what solves their problem is having a local intranet. Don't host it on the cloud, just host it locally.

Yeah, yeah. So I should have clarified this. In this case by edge, I mean on-premises. So I would bring a server to them that can run all this system and that can run some of the open source or even licensed LLMs. I need just text to speech for the audio notes and then I need a larger LLM for classifying the notes. Probably we're going to build our own at some point and make a smaller version so we don't need so much memory.

So we run just to run those locally. There is a bit more classic edge than probably that's what most people refer to edge where Already on the device that we placed on the machine, we will do some pre-processing of the data because for example, instead of taking raw data from the sensor, from the accelerometer time domain, we will maybe run an FFT there, extract some features and then push those to the local server.

Yeah, rather than full waveform. Yeah, because mainly it's not just for security because at this point you're running everything locally. It doesn't matter if you set all your system right.

It's super secure. just too much data. It's like bandwidth.

Yeah, I've been I've been recording Now since December with one customer on like 10 machines and every day is 30 gigabytes of data. Like, or sorry, one month, for machine is 30 gigabytes of data and i try to compress it you know keep it lean but this is way too much so that's why for me i think there needs to be some pre-compute at the edge um at the true edge right where the sensor is yeah that makes sense yes we uh We built safety sensors and so our systems are Vision based web cameras.

and we have a Okay.

A product line and extension of our safety product that does monitoring is we have a color camera. We've got 3D data and we have a product line. Processed. Versions of those that mean cracking data. Different things we want to do with it, like for example, how often does a user approach the robots? So typically how our sensors are used, they protect a robot. You get rid of the robot fence and you just have 3D safety sensors around it.

So some of the things they want to know are, how often is the robot running? So similar to what you're doing there.

How often does it run? LUCAS RIEGEL: The spindle time, but the equivalent of the robot, yeah.

How long does it turn off? When people approach it, how long is it off? What's the downtime of that system? Um, And and really kind of Importantly, why did it go down? Was it an improved user actions, like somebody bringing in a pallet or did somebody just walk through the safety field or open the gate? What was going on? They don't really... They don't really know right now. They have a hard time trying to tie together the safety event and the visual.

And so anyway, that's one of the products we offer is kind of an analytics solution on top of that that uses our camera data to provide that information to the user customer. And one of the things we ran into is the band thing, right? You... We have early FPS images, you know, and It's a lot of data.

It's a lot of data.

And so it's a good one. you can compress it for sure. Image compression is easy to do.

But then you still are sending JPEG streams over there, ethernet or Wi-Fi or to some local-You still need to process it afterwards. You're not done.

Yeah. Yeah. And so we have similar issues there. and So edge processing is like we use a-In NVIDIA, like Jetson computer, and it does, it can do a lot of the compression, but it can also do some processing, like can we extract How much information can we extract before we send the video down, right? And then... kind of features or like, you know, we can do pose detection so we can figure out like what, how a person was positioned when they approached the cell and where they were.

And we can send them, you know, a small sparse point cloud down instead of the entire image.

Yeah, yeah. And things like that. Do you run some models? Do you build some AI models to run on that?

We don't. We, um... We haven't built any ourselves.

The stuff we're doing is simple enough that we can reuse existing models, pose the tensions.

Yeah, probably NVIDIA offers them already, right?

Yeah, they have some that work on the Jetson and the timing. I don't even really think of them as models almost. I mean, they are.

They are, but it's more like the classic ones before the Chagipiti. They existed 10 years ago or at least five for sure. Yeah, it's stuff without heritage. I mean, those are good. They just work. So you don't need anything else. Yeah, interesting because we were... So now I'm trying to figure out what... should be my local server that I run and I was looking at some of the computers that Nvidia offers they have this DGX that is like 5k that seems pretty interesting and yeah Then there was another one more like for 2000.

I forgot the name. That's more the previous generation that they had. It was this Jetson box, I don't remember how it's called, but yeah, it was a small, you run less, less like, demanding models but it's still pretty good it's just a tiny box that you could put in the rack Yeah.

Those systems are very capable. Even the small ones.

for images like there. Was it the AGX? Orin was kind of the...

Yeah, I think the Warren. Yeah, yeah, that one. Because for you, you need one per robot, right?

You need to put that on the edge there with a robot because that one, just a single robot generates a lot of images that you need to. Yeah, yeah. So right now our sensor is...

And that's the compute for the... I'm just compute, but the... It runs the cameras, does the processing, spits out the data, does all of it. It's a first sensor.

Then what happens if you have 10 robots in that facility? Does that data spit out to your cloud or to a local server and then goes to the cloud? There's some sort of gateway.

Yeah, usually...

The customer typically wants on-prem, at least the ones we work with. They're big companies typically right now. Um, Smaller companies, you know, they don't necessarily, they don't really want more equipment. They don't have the privacy concerns half the time. And, you know, same thing here. Basically, it's like... Can you do it on the cloud and That's fine.

That's heavy for us. But the software is the same, right?

It's just the Python.

Nice, nice. Yeah, myself I'm trying to set up gateways, you know, where I have like, 'cause in my case I only have usually one sensor per machine, maybe two, because we might have to put a, the mic in the front where the operator can talk and and accelerometer closer to where the vibration happens, maybe on the back or inside. So we are trying to figure out different options. So you'll have maybe two sensors per machine.

Then you can imagine there is like a factory like 20-30 machines so you have all those deployed all over So then we have a central gateway that receives this data and then either it sends it to the local on-premise server or to the cloud deployment. So that's how we architect the things for now. Um... Yeah. Who? Cool. Yeah, so I'm curious if you have seen with your customers in your domain the need or if there is a gap to sense more like vibrations or like gather data from operators like things we are doing.

Do you see something missing? in that area Um I don't know.

I mean, vibration sensors have been around for quite a while, right? They use those for predictive maintenance all the time. Uh... The user side, I don't know. I have not really paid that much attention to that part of it. That's an interesting idea. Because capturing that knowledge and that... Good notes. I'm not sure how they do that right now.

Yeah. Yeah. I mean, it's definitely valuable information.

At least some machine shops that I work with, they have this software. At least one of the popular ones is ProShop. that they have them run the jobs like when they print like a part so they keep track of that through there and there's like some fields where you can input some notes but you know it's kind of ancient technology it's like it's not the most user-friendly interface. And nobody does anything with that data also.

There is no like...

processing and seeing okay those notes are this type of notes, these are other type of notes, those are issues. There's no easy way to put them together and give an overview of what's going on. Um... Yeah.

Well, it seems like a great...

source of data anyway, if you can get it all.

Yeah, and then our idea is also to hopefully use this data to train new people because there's a lot of shortage in factories of skilled people and a lot of people are retiring. So if factories manage to capture those data before people leave for a reason or another, or maybe even if the experts, if they're on vacation or they're not in that weekend shift or something, you don't want to call them, you could just ask your factory AI brain, There you go.

So that's how we want to go about it. Um... Yeah, the other thing I wanted to ask you since, but okay, maybe you're building something different in robotics, but I'm doing a bit of research, so like besides my startup, I have an appointment at the university as a research professor. that pays the bills uh like part of that and we are doing research on like sensors like i'm continuing my research from my phd on ultrasonic sensors like those miniaturized ultrasonic sensors that fit in a tiny chip i'm not sure if you're familiar if you're familiar with them they're called the p-mats piezoelectric micromachine ultrasonic transducers oh yeah yeah yeah yeah okay so actually with the pia gianluca piazza cmu that year that was there we were doing some some of that with him.

and We are looking at some application, at least I'm looking at some applications where You could Actually, one step back. The classic application is time of flight. So you speed out a wave like an ultrasonic, when it comes back you check the duration, right? And then you know the distance. The typical sensor you have on the market is like one of them is from Chirps InvenSense and it's at high low frequency so big wave and it can go far.

So you can, it's okay. It works well if things are like at centimeters or meters, like several centimeters, but it doesn't work well short range. So I was trying to see, I was reading a bit around but you know I haven't heard from people working on this necessarily that when the robots try to take something to move some part and have a grip on the thing, a hold of an object usually doesn't know, it doesn't have a feeling of like how strong it's catching or if it's close enough.

And most solutions now are camera based, right? And maybe, and that requires a lot of processing. And also if things are hidden a bit out of the field of view, maybe then camera won't work. So we are trying to understand if we can build those sensors At much higher frequency, they could attack much shorter ranges. And not only, it could detect touch as well. Okay, now it's made contact and potentially also pressure, how much pressure you apply.

You might need two versions of this sensor, but it can do both. It might not be able to do both in the same chip, but you can do two chips that do next to each other, right? So I was trying to understand like the need for that in robotics either for a regular grip How would you call them? It has the two sides, the clamp versus a humanoid that has all the fingers.

So you would put one per finger.

Have you seen issues with that? Do you know what people are usually using? What type of sensor is right now?

Yeah, so some, companies use forced-torque sensors for They're kind of relatively big, right?

Some people use...

Right. There's elastic actuators where they can measure the current.

And they can kind of get a source of the feedback.

or how much forces in the plastic part of their actuator. I don't... I'm not a... It's been a long time since I looked at this area. But...

I know there's a lot of different methods out there.

I don't really know what. companies are using. I mean, honestly, mostly what we see in in automation. is like they're vacuum cups that's the way that a lot of things are picked up is with a vacuum cup you can't put that in a humanoid robot for sure yeah you can't put that on a humanoid uh I don't know how they're doing like the fingers and It could just be looking at the currents, right? And then training some big model on the current.

Maybe that works better than it used to. I know that there's, that's a whole field, right?

Yeah.

Look, it's noisy motor current and figuring out how much force you're applying on the things. I don't know what they're doing. I used to be in that area, but I have no idea what they're doing anymore. 15 years ago, I did some stuff in that.

Yeah, I'm not a robotics person, but I'm robotics curious and I'm more on the sensor side. I'm a MEMS guy and I'm curious if we could... you know, if it's worth exploring more that area and trying to come up with a more compact solution, not necessarily low power, but compact in form factor, it's something that could be interesting. So I'm trying to talk with more people and ask them about this that are more in this field.

Yeah.

Um... Hmm.

I think of people and feel.

Have you talked to anybody at Picnic Robotics?

Picnic Robotics? No.

Is that an association? They're more of a software company. That's funny. They do...

I don't know. They do some stuff where... It's like a, I think the main product is like a visualization tool for robotics, like pick and place applications.

Yeah, I'm looking now on their website. I'm not sure.

I see the suction cup.

Yeah, I'm sure there's better people doing that.

But uh... Where are they based?

I think they're in Colorado.

Colorado. Are you in contact with anybody there?

Ah. I haven't popped in for a long time.

Oh, it's a big team, nice How many people are you at your company?

We've got, we're small. We have three people.

Nice. I love it. I'm alone now.

Yeah. Well, you'll get there. It took me a while to get some more than one.

I know. I know. I'm struggling. I'm at that point. It's like, it's like a, I don't know how to call it. I got a bottleneck. I feel you're choked here. Like, you know, when you choke the engine, you have this old car, you put something on the escape on the back and it cannot go. So that's where I am at.

You need investor money to hire people. To hire people, you need to convince them and need money, right? And then to advance the product, you need all the above.

Oh, yeah. Endless chicken and egg problem.

Yeah, catch 22, but it's worse. Yeah, both.

So I'm glad now that I have, you know, The AI programmers.

Oh man. It's a life changer.

Yeah, it's amazing because before I was scared to take more like big tasks because they're just a pain. Like all this infrastructure to build Cloud Edge, all the things. I can build many parts of it, but it's just you need to debug so much. It's always how do you write this line and you cannot have it all.

My local memory is limited.

Yeah, Claude the Code and I have developed more software in the last two months than I did in the previous two years.

It's same, yes, it's incredible.

The best thing about it is before when I was working on a project, It's like, okay, I got to block out an entire day to get anything done. And because it's just context switching was so terrible.

Yeah, it's the context switch.

You need to upload everything in your mind and you cannot be distracted. Yeah.

Yeah, and so it's like that stalled me a lot because it's like, well, I don't have, I can maybe work on this for two hours, which isn't enough. So I'm going to get a start. But now it's like, I can sit there, you know, in the evenings or all day. I can just, like Claude's actually, Claude is working on something right now for me. I've got like 10 agents booted up working on something.

It's a...

It's really amazing.

What? Thank you. So, a little scary, really. What are my kids going to do?

Well, they're going to do architecture, more like high level. I feel like writing software has become... a bit too glamorized but And then I'm hearing the Bayer and the software engineers, they make so much money that I feel like it's ridiculous. And now I think they... kill their jobs, I think.

Which is fine. I'm a hardware guy, so I don't care.

Yeah, yeah.

Too bad, guys. Sorry.

I write software just for fun and because I need it for the hardware. Yeah. Yeah. Yeah. Yeah, all right. I won't keep you too much longer. It was really nice meeting you and learning more about what you guys are doing. And it was good sharing some thoughts here.

Yeah, absolutely.

Yeah.

Let me know if you want to chat more in the future or if you think whenever, you know, talk, do your thing at the company, if you... If you think anything could be interesting for my company, just ping me.

Yeah, I'll keep you in mind.

I'll send you the website here of our company.

Sure, please. And by our, I mean my company. Uh... So it's abonito.ai. And we have a little, I mean, I have a little bit of like information there. Things are evolving continuously, obviously, as I talk with the customers and we try new pilot things. Yeah, that's about it.

Great. Website looks very nice. Thank you.

Thank you, Claude.

I was just going to ask, is that code, like, I need to read you ours if that's what Claude does. Yeah.

So, basically, back in the day, like, 15 years ago when I was in Italy as an undergrad, I used to do websites for, actually, in high school, actually, started to do websites for, like, you know, side gigs. They would pay pretty well. And it was before, you know, any framework existed. WordPress was just coming out, you know. So, yeah, so I had to do everything from scratch with my PHP, MySQL, all that from pure scratch.

Then more like a few years ago, maybe five years ago, I started getting to all these new fancy frameworks like TypeScript and all those React, all these libraries on top, which was, oh, my God, that was revolutionary, game changing. I could make so much, so nice websites so quickly.

And then I reached Cloud. about it and it does it for you so Yeah.

I might have to take another spin on my website when I get a chance. I use Wix and I did everything...

You don't need any platform anymore. Yeah. Cool. This is going to just do it for you.

And it does backend very well as well, handling database and all that. So users, pretty good. Yeah.

What Quad is working on for me right now is a requirements management system for like a project. I didn't know anything about this until I started developing safety products. But doing projects and speccing out requirements, like what the project or what the product needs to do and any standards it needs to follow and all that.

And...

So it's building a web app for me right now, that does all this. And I could, I tried a few out online and like any ones that were good were like, you know, $200 to $500 a license, and you had to have a minimum of, per month, you had to have a minimum of like 10 licenses.

Yeah, just pay Cloud Max, right? Pay Cloud Max and it does everything.

Yeah, and so I'm like this So, you're using the agents while we are on the call.

So how do you instruct that to Cloud? Yeah.

From command line, from VS code or just.

Yeah.

Yeah. I use, I use the cloud plugin on VS code. Yeah. And then... I mean, each Asian is like a skill. It's like a markdown file. I don't know how you even code in... AI stuff. You used to have to say, you're an expert in XYZ, you've done this, this, you give it a background and a history and then your job is to this, right? And, well, that's what an agent is, is you do that and you define it in a file, but then every time you want that agent to run something, it's like a forward slash or a backslash and then you type that in and then you give it the body text and then it'll start working on it.

But you can do more than that. You can have teams So I've got, like right now I've got a team of like six AI agents that are, you know, one is a business safety product. So one's a, a safety lead.

One is an art, like a, a standards reviewer that works like a certification agency.

One is the MECI. One's an electrical, one's an FPGA guy. And then, uh, And they each have their own jobs. And then you can define like a group of these agents and call it a team. And then you can spin up the team and have all these different agents look at the problem from their perspective.

And then...

and it's a I'm not sure I've perfected it yet, but it's super helpful. We were a tiny little team and I was usually doing these perspectives anyway. It'd be like one at a time though. And I'm like, okay, well, Like here, now look at it from the MechEase perspective, what we need to think about. But now it'll just do it all itself and then...

Work straight. I don't know, there's a whole, there's YouTube videos and YouTube videos, you know, on these things and, It is.

It's pretty helpful to get things done. And it's not perfect, but boy. I don't think that.

I mean, it's like a, it's like a, at least a 10 to 20 X in like productivity improvement using, Yeah, I haven't looked that much to build a team of agents because the way I'm using cloud is basically I have, for example, for this project, for the platform, I have my fleet, I have the platform that the user sees, I have my gateway code. So I have a VS Code project open with several repos. So that's my workspace.

And then I just chat with Claude there and I say what it needs to do. Each repo has a Cloud MD file that it knows about the project was supposed to do. butter It's always one agent at a time and usually finishes pretty fast what I need to ask. But I never use this fleet of agents that I have them and reuse them kind of a team that's I like that how you like a teamYeah, I started doing it about two weeks ago.

So I used to do it the way you do it. I still do, but man, is it, Yeah, I need to figure out how to do it.

I'll look at the YouTube video.

It's interesting. Yeah, and...

Because then I can tell him, look, I have a team of five agents.

Yeah. It's...

Yeah, I... The bad thing about these AI coding tools... I use it for everything, not just coding. I use it for market research. I use it for sales, all that stuff. But... The thing about it now is I can move so fast. I used to have all these ideas and I never had time to do anything with them.

But now I have all these ideas. And then I can do it. And then I can manage it.

The problem is I stay up too late now. I stay up until like, you know, 1 a.m. and then wake up at 5 because I can't stop thinking about these different businesses.

I see. I see the problem. Okay.

Oh, that's pretty cool stuff. That's for sure.

That's cool. Yeah, I'll definitely try the team thing because, yeah, I want to build a sales team. I want to build an E team, ME team, all that.

Because now it's just different files. It's not different hires.

Yeah, yeah. It's a...

Yeah, I mean, all you gotta do is say, Claude, like... How these guys work together, kind of.

Yeah.

One of the things is having them have memories Is, uh... Right. That's helpful. A lot of times, like, you boot them up again the next time and they don't remember what they did.

Yeah. Right? But if you say your memory is this file...

And then they write to it after everything they do. Then you got to be careful not to make it too big, just like your CloudMD file.

Yeah, you can probably have it like...

have different files that may be in one file per month or one file per week or something like that. So they don't have to reference the whole year. But just reference. Oh, reference only the last week unless otherwise told. Yeah.

Yeah.

Yeah. Anyway, it's cool stuff. That's a huge productivity booster for me anyway.

Yeah. Good, good knowing how, how you are doing it because yeah, it's interesting learning because this is new for everybody and it's good to know what works and what doesn't. Yeah. Yeah, cool. Okay. Well, uh, you know, if I think of any way you should talk to, I'll let you know. Thank you so much. There's stuff float around in my head. Uh, I just haven't, I need to think about it a little bit. Yeah, for sure. For sure. And maybe we can catch up in a month or two whenever you have time. All right. Good to meet you. And thank you. Thanks a lot. All right. Yeah. Thank you. Sorry.

My camera wasn't working. I got to figure that out too. The green lights on, but it's just Kim.