UFB1

Interactions 1

AbnitoMeeting with Vitaly BulatovMay 6
Vitaly BulatovUFB

Meeting scheduled over his google scheduling page.

Summary

Summary

Flavius (Abnito) and Vitaly (UFB) met to explore potential synergies between Abnito's machine monitoring/note-taking platform and UFB's humanoid robotics business. The conversation was exploratory — Flavius demoed the Abnito platform, and they identified some relevant use cases for UFB's robot fleet, though UFB is not a direct target customer. Vitaly offered to help spread awareness of Abnito within the robotics industry.

Key points

  • Flavius recently got a technical co-founder, allowing him to shift focus to sales and partnerships
  • Abnito's platform combines a physical IMU device (Bosch) + audio note-taking app, with AI layers for transcription, machine classification, and categorizing notes into buckets (issues, setup, maintenance, etc.)
  • Target verticals: machine shops (easier entry) and semiconductor/research fabs (higher value, harder to sell)
  • Current customer: a Bay Area machine shop with 13 machines on annual contract; use case included sending maintenance notes to their CNC servicing company and passing ISO 9001 audits
  • New direction: phone-based note-taking without the physical device, lowering the barrier to onboarding new customers
  • Platform is now dockerized/open-source-stack, enabling on-prem deployment for customers with data residency requirements (fabs, enterprise)
  • Vitaly noted UFB collects IMU data from humanoid robots but hasn't processed it yet; interested in predictive maintenance use cases
  • Spindle time detection from accelerometer data (distinguishing idle vs. active vs. setup) discussed as a compelling feature
  • Flavius offered UFB free accounts (up to ~5 users, ~10+ robots) to trial the platform

Decisions

  • Flavius will give UFB free accounts on the new Abnito platform to trial with their robot fleet
  • Flavius will prepare the enrollment/onboarding system before sharing access

Action items

  • Flavius — Get enrollment system ready and share free accounts with Vitaly/UFB team
  • Flavius — Share the IMU feature extraction parameters/methodology with Vitaly once ready
  • Vitaly — Onboard UFB's robot fleet on Abnito and explore use cases; help spread word about Abnito in the robotics industry

Follow-ups / open questions

  • Can Abnito's platform create meaningful value for a robotics fleet (vs. machine shop)? Trial will test this
  • Is there a viable go-to-market angle for Abnito targeting robotics/robot fleet operators?
  • Spindle-time-equivalent feature for robots (idle vs. active vs. setup) — potentially interesting but UFB doesn't yet have standardized operations
  • Both founders are meeting again at a Bosch reception the same evening — may continue discussion there

Transcript

[Flavius, founder of Abnito, the platform product] [Vitaly, founder of UFB] Hey, sorry. Been good, been good. Nice. Also pretty good. Yeah. Just applied to IC yesterday. Oh, man. Good luck. Yeah. Thank you. Thank you. Yeah, I managed to get a co-founder in the past few weeks, so that's pretty good now. Oh, man. Wow. Were you looking for a technical or like... Yeah. Yeah, tell me more. All is technical. Nice, yeah. Yeah, yeah. I need to hand over... Hopefully, most of the technical part so I can focus on more customer development and finding partnerships and trying to sell these products. That's what they need to do. Otherwise, business doesn't go on. Well, yeah, yeah, I'm Tell me more like Do you have a So like already a particular customer segment, is it particular machine you're targeting? Yeah, yeah. So now the segment, the vertical we are looking at are machine shops mostly. because machine shops are easier to talk to, easier to get into. They have less... stringent requirements, at least some of them, you know, the bigger ones probably that they're more enterprise kind of vibes where they need some security and they need to You need to be maybe even bigger startup. You need to prove to them some stuff. And the other vertical, which is also tougher, it's like semiconductor fabs. And in particular, more like research fabs, government fabs, university fabs, which are a lot of them in the US. So I'm not talking about those modern TSMC fabs, because those are already pretty automated. It's more like the old school fabs, which plenty of them in the US. Because they are bigger, you know, compared to machine shops, they can probably take they're worth 10x in terms of like what they can spend and how much value you can bring them Yeah, we have a customer here in the Bay Area, a machine shop that is using our devices across the whole machine shop. So that's pretty good. They're a small one, they have 13 machines, so that's how many devices we deploy with them. And now we're going to give them the option actually to use also just from the phone. So... Oh, so you're like you attach the phone to the machine? So now, so... Let me show you. So So now how it works so far is that you attach a device like this to the machine. So you attach this, you take an audio note, so it's just like a note taker, and it sends vibration. So that's what we have so far. But the platform where people see all this and It's a website. And vibration is an IMU, right? It's an IMU. Yeah, it's the Bosch IMU. The same you guys are using. Yeah. So now I just made the website more mobile-friendly. And I gave an option here at the top to just record a note from the phone. Of course, this doesn't sense the vibration, but this solves most of the problems because The first problem we are solving is that we help them digitize all their notes and This type of noise can be like, okay, the machine broke down, there was an issue, this is the error code. or I'm setting up this job with this many parts for this customer. Or I'm doing routine maintenance and I'm changing the oil and adding coolant to the CNC machine and things like that. And so that's just like an operator just makes a note of it. And right now they do make a note of that just from the device itself. Exactly. Exactly. But we noticed that, well, maybe... Maybe some use cases, at least for some machine shops where All operators have a phone, then they can do it from the phone. But maybe not just operators can do this. Maybe floor manager can walk over and supervise. And when they see issues, they take notes on the phone. And the same way, like machine shop owners, like the owner, I said, oh, I wish I had this on my desktop where I take a note from the desktop directly because she goes back to her desk, she opens the dashboard, sees what's going on. And she said she would like to see from the phone and take notes from the browser. So I said, why not? That's easy to do. And this gave us the idea, okay, maybe, To deploy this, we don't need the device initially. You first just give the app. And see how that works and then once they like it then you give them the devices as well so kind of easier to get the foot on the door even just share the app with many people even people that are not in the manufacturing industry necessarily like i can give you an account just use it for free so yeah if you guys might find use case or like just find bugs you know i was like oh shit this doesn't work yeah and i mean uhYes. Actually, you might. It's interesting because in a way, with a device, you're particularly targeting machines because you can do some predictive maintenance on vibration data and stuff like that. In our case, we deal with humanoid robots, so they do supply some data. They do have the Bosch IMU. And you put your own IMU, so you can do this as well. Yeah. But still, like, I'll... Product is not management of a fleet or you know our product is deploying that fleet so It's interesting to generally see how... I mean, you can be... Yes, I like the app idea. and testing it out. I can show you, I can share my screen just quickly to show you. Yes, we do like for example a particular use case like We do go to a lot of events with different robots different people take them out and at some points of time Robots get out of calibration and they start making like a cranky noise sometimes in the actuator So I get it take it out. Yeah, and yeah take it out exactly right so because it's it's it's not a I mean, I guess we can capture it from IMU somewhat, But still, yes, taking out is literally what we need. So our idea is that at least from the operator when they have the device is that they can, okay, now they take all those nodes. And if there was a signal from the IMU, we can correlate the issue node, the weird noise kind of with IMU to see if we could have predicted that. Or at least... re-understand that because maybe you notice that crank and you know what's going on, but maybe you get a new employee And then they hear the same sound, but they don't know what it means. So you could flag it for them. So you can correlate that. But even without IMU, the idea of this app here, you can see the screen, right? Yep. Okay. So yeah, it's just to collect notes and put them in different buckets. So now the way it works, first thing, so this is a demo account that I gave to YC, so I'm just going to show you. Like you collect the audio notes, you can re-listen to the audio, you You get through AI, you get through these AI models, you get the transcription. That's step one. Then there is another AI layer that understands what machine are you talking about or what device, whatever system you have. We just call them machines, because machine shops in this case. But it can be like, oh, you just put 10 robots here or like 10 of other things. You don't even need this necessarily, but it's a thing for manufacturing. Usually they do have systems or something they work with and they want to monitor or take notes about. So the AI understands where... because basically we have a classification layer for that. And the reason we do that because, okay, initially I link those devices when I put them, I link them to the machine. So it doesn't need to know. It goes directly to the machine's bucket. But now that we can use the app and you just take a note from the phone, well, what machine did you talk about? Well, you can just add that layer that understands and puts in the right bucket. And then the other bucket is that you can put in this issue set up, shift send off, or other buckets you decide. And we just have some options here that Well, we can generate those buckets. We can precede them for customers or the customers can just decide their own buckets here. And basically the AI takes in all these buckets with their description, the note, the transcription, and then it decides, okay, it's in one bucket or another. This is very simple AI usage, right? I guess also, like you mentioned on a previous step, like how would you know which machines? In our case, yes, that could be, because we have 20 robots and like you put them all here taking them so yeah like we can have it we do have a name for them actually so yes i guess uh it's the this step before onboarding step i guess would be useful to say like yeah when you do notes you should mention then like number of the robots like yeah yeah or something like that so here It takes in consideration all the description you put here. And we can add different categories. So it has name, serial number, the locations. Maybe robots.free doesn't make sense. Location, you leave that empty. Some description. So it takes all this in. at classification time. So even if you don't say correctly, oh, CNC machine center one, but maybe you say some description or just say serial number, it will still understand. Yeah. Then location can actually be helpful even though it's not static. It's actually useful to know where where okay Then, of course, if something goes wrong, you can modify and reassign the node to something different. You can just change either the category or the machine or the robot, the date and time. You can add manual nodes if you want to. And then you can... Oh, one interesting thing. Okay, let's say you're a company and you have multiple users. So here I only have one user for this company. You might know this name from the sitcom. Silicon Valley sitcom. Yeah, yeah, yeah. But if, yeah, I gave it here an option to select where did the note come from? Did it come from Richard? Did it come from another employee? Or did it come from one device that is installed? So, for example, these are notes from the device, but this is notes from Richard. And then if there's another employee, it will be notes from another employee. So there is a little sign here, Richard Hendricks. And when you get the data from the device itself, how much do you do now with the MU data and predictive maintenance? Honestly, I'm just gathering data. So I... what I did so far is that I have some local processing here because to beam out all the data is too much. It's like terabytes of data. I don't want to do that. So I read some papers about predictive maintenance and I extracted classic features that people extract. Because most of it is noise. You look at the spectrum, it's like nothing. So you just extract some RMS values, this thing called kurtosis, and a few other things. I can share with you what we extract. But honestly, I haven't. time to process and make it useful for the customer yet. So for now it's mostly about the notes. Yeah. And one thing our customer, we have one customer of annual contract. What they did, the interesting use case for them was to take this notes, they took screenshots of the notes when there was an issue and they sent for service for their machine. There is a big company here for machine shops that they do distribution of CNC machines and servicing them as well. And we talk with them and they could be an interesting distribution channel actually for us to integrate directly with their system where a customer can just push a button to send a request for maintenance instead of just taking screenshots and doing all these things. So this came directly for the customer. It's pretty exciting to see them. using the platform like this. And yeah, they could be a distribution channel for us. Yeah, that clearly creates a reason for me why I make these notes. Yeah. That's nice. Yeah. And we build also audit report for them. Like from all this, for example, in the bucket, where is it? Do I have that bucket now here? Sorry. Well, I don't have as an example here, but we put one bucket called routine. This is all routine maintenance that they do to machines. And this is requested of machine shops and other factories to do some routine maintenance to all their equipment. And this goes in not predictive maintenance, but preventive maintenance. So if they do all this preventive maintenance, it means that they fall under a certain quality management thing and they can get the ISO 9001 certification. So they have audits routinely. So we build a report for them and they pass the audit. The auditor is happy with it, with our solution. So would that be a plugin or something to make me ISO certified or something? Like, cause I would be appreciative, like, yeah, cause it's a specific steps, right? I would need to know, like, yes, it's a preventive maintenance that I need to do. I also need to do like regular checks, right? It's like the list of things I need to do. So will there be like a page for me? Like, yeah. or schedule or something to me remember to make those notes to be ISO compliant? Yeah, so I did import it here in the new platform because I have the old platform in AWS where the customer is. We have another page, here, like on the menu, that it's called just checklist. And they have a checklist with tasks and a frequency. And then they printed that and put it on the machine. They can do that on a Word document, but we just made it easy for them on the website. And the ISO certificate... So they already had the ISO certifications. It's just that they need to pass the audit every year and they need to show documentation of it. The way they did it in the past is that they had the notebook for every machine and then... regularly the machine shop owner On a rubble gold, take those notebooks, put the notebooks in an Excel sheet. So it's all very manual. And then the operators, they don't take as many notes because they have to write and stuff. So hopefully audio, it's easier. It just creates less friction and we can automate all this. Yeah, that's really nice. I mean, we are kind of... I mean first I guess yeah, I'm very interested because like I've been dealing with hardware and machine shops for quite some time We are probably not as direct customer because we don't have a machine shop. We are more of an interesting exploration because we have a fleet of machines, but you know Quite different so we'd love to explore especially the airport now Also, yes like Bosch collaboration, anything we have to do. And I mean, primary thing that we are like our product is essentially attention around robotics industry. So we would love to kind of just help with spreading the word. Yeah. Thank you. That this exists and yeah. Yeah, thank you. And I really appreciate it. I think anything to spread the word and show other people how this works and find new use cases is just awesome. And I'd like to give this for you, a free account, just to give You can onboard, I don't know, 10 robots, whatever, how many robots you need. You can get a few accounts. I don't know, you and somebody else at the company, maybe like, I don't know, five accounts. It doesn't matter to me. And you can just use it for free. See how you like, see how much value it produces to you. And then I can go to other platforms companies like yours to sell it. Yeah, yeah. I just want to see if it creates value to you as well. Because to Machineshop, it seems that they create value. Machineshop is a really great product. And I mean, it's an interesting thing because we do have machines that are somewhat different, but yeah. I think it... Yeah, if this could work for robots and robotic companies, I mean, that could be huge because it's a big topic now. So it'll make investors happy as well. Yeah, yeah. So let me get the kind of enrollment system a bit ready to share with people because this is a new platform I made. We were in AWS before I was saying and we got out of AWS and now it's all open source tools. And this way we can deploy it on-prem as well. Because it seems that many customers like fabs and bigger machine shops they want on-prem they don't want the data to go out so for this type of customers we want to just be able to deploy this easily just another push it's all dockerized so you can just build it easily that's nice actually um Will you be at the Bosch reception today as well? Yeah, yeah. I'll be there tonight. Are you guys going to be there? Yeah. Awesome. Awesome. Yeah. So yeah, we'll definitely see each other more and yes, like send me the accounts and I'll more importantly think more of Both machine shops and the area and generally, like... about the platform. Because yeah, it's a pretty cool product. Thank you. I don't know how I can help on my side for you guys, if I can be of any help or like if any of our technology in the platform, if you see anything that is interesting, like, oh, how did you build this service or whatever, I'm happy to share or anything. I don't know how I can be helpful now, but feel free to ask me any questions. What we are at right now, we've solved it is the Yes, exploring data from the robot to understand some way of predictive maintenance for those because they just hit the market. Like nobody has the data on them. Nobody has the data of how they break. Just stuff. So that's kind of the reason behind what we are doing. Like it's all fun and attention, but we are like running hours and hours of breaking humanoid robots and seeing what goes on. So Where do you put the data now? We just, I mean, AWS. But like in the sense, like do you just accumulate the Axler IMU data somewhere and for now you haven't looked at it? Yeah. I mean, yeah, we have it, but yeah, it's, I mean... Yeah, it's a lot of data. Yeah, yeah. Yeah, maybe... I'll show you once I have something ready on that side, show you what kind of... parameters I'm looking at, I call it features, I just extract something from the signal, this way I don't have all the data there. Yeah, I'm trying to figure out how to make it useful for the customer. Yeah, one thing that was interesting, one customer told, I showed them a preview of some data. It's like, oh, look, I can tell you how many hours your machine was working. And it's like, oh, is that spindle time? so what's been the time spend the time was that so basically they want to know exactly how many hours the spindle runs, because that means it's running a job. Because the machine can be on, but not running. This means that it's idle, it's working, but maybe either the operator is not good enough to set up the job and it's taking them too long of a time. So they're asking, hey, can you distinguish between setup time and... active going Because they want to know, is it idle or is it idle because the operator is setting up? And then if I know spindle time from the accelerometer, and then the operator can take an audio note and say, hey, I'm setting up. They don't even need to say I finish setting up because you'll see just the vibration that I'm going crazy up And then you can distinguish. Okay, we can say oh this operator that this much efficient that's setting up jobs. These are not So yeah, that's one feature. I mean, that's a really cool feature Yes, I mean our operations are not yet as then like we don't really have any standard operations. That's kind of a thing but Yeah, I guess we do idling of robots as well. Yeah. I don't know if that's something of interest, but to them that spindle time was interesting and I'm trying to make it a feature again. Thanks for sharing that. Yeah. Right. Cool, man. Good to talk to you. And I guess I'll see you tonight. Yeah, see you. Thank you. Awesome. And thanks again for planning to invite us to the event to see the fight. Oh, yes, definitely. Yeah, awesome. See you. Okay, see you later. Bye-bye.