He shared some links to their model, I can find on website. Plus YouTube video.

---

Summary

Summary

Flavius (Abnito) caught up with Pete Warden (Moonshine AI) to share progress on the factory IoT/voice AI pivot, discuss technical architecture, and seek fundraising advice. Flavius has his first paying customer (a machine shop) converting to an annual contract and is pursuing partnerships with Intrinsic and Bosch. Pete offered advice on fundraising strategy and agreed to make some introductions.

Key points

• Flavius closed first paid pilot with a machine shop (~10 machines), now converting to annual contract; customer is actively referring him to others including a larger shop with ~50 machines
• Flavius met Intrinsic (Google robotics) through the pilot customer; they expressed partnership interest as it's "not in their turf"
• Flavius migrated hardware from Raspberry Pi to ESP32, cutting BOM from ~$60 to ~$20; built own fleet management on AWS IoT replacing expensive Balena OS
• Bosch partnership in place for high-frequency accelerometer access; Bosch Italy team helping connect to European factory customers
• Flavius has 3+ months of vibration data; currently able to detect machine utilization/spindle time but not yet predictive failure signals
• Core value proposition: voice notes from operators → stored knowledge base → compliance/ISO audit reports + tribal knowledge chat assistant for training new workers
• Pete shared the Raspberry Pi doc chatbot project (hypothetical question generation + sentence embedding matching for documentation Q&A); discussed relevance for machine manuals on edge
• Pete shared Panasonic/Synaptics dishwasher demo — on-device FAQ lookup running on a $5 chip; limitation was manual FAQ creation, now solved with automated approach
• Pete cautioned that predictive maintenance tends to be very bespoke and hard to scale — has seen multiple startups fail on that path
• Flavius had a YC interview in January; rejected mainly due to solo founder, small team, and hardware scaling concerns
• Fundraising target: ~$500K pre-seed (prior angel round was $100K)

Decisions

• Pete will send introductions to potential angel investors; Flavius to send a 2–3 sentence blurb
• Flavius will reach out to James at Bloomberg Beta again (previously passed at medical device stage)

Action items

• Flavius — send Pete a 2–3 sentence summary of Abnito to forward to potential angels
• Pete — identify and circulate Flavius's blurb to relevant angel contacts
• Flavius — re-contact James at Bloomberg Beta
• Flavius — look into AngelList; treat raise as rolling closes (~$25–50K increments) rather than one institutional round

Follow-ups / open questions

• Whether Moonshine's edge ASR/embedding tech could be integrated into Abnito's device once edge deployment becomes a customer requirement (flagged as future opportunity)
• Intrinsic partnership — next steps and fit unclear; conversation scheduled but no details
• Bosch Ventures flagged as Series A target, not relevant yet — when to re-engage?
• Flavius's Plug and Play desk contract ends June — looking for more vibrant coworking space in Mountain View; Pete's building may have desk rental but details unclear

Transcript

Hey Pete, how are you? Good, how are you doing? Pretty good. Uh, give me one sec. IFor some reason, myEarpods didn't connect, ah. Let me press on them again. No worries. Sorry. Hear me now? Yes. All right. Actually, I cannot hear you. Okay. Yeah, whereabouts are you? Sorry. Whereabouts are you?

Oh yeah, so I am at Plug and Play.

Here's where I have the desk. Yeah. Ah, yes. Yeah. So, I've been hanging out here. Not a lot of people here. I think they're gonna have a batch at some point of some startups so maybe it will like populate but yeah yeah so It's good it's silent here. How about you? So I'm actually working from home today.

Oh, okay. Nice, nice.

Yeah.

So do you have an arrangement with Plug and Play just out of curiosity?

Yeah, so far I'm just paying for the desk, but I hope to be able to get into their accelerator. So that gives you three months free of desk. Yeah, but they don't necessarily invest on that program, but also they don't take anything. It's like for free. So now since we did the pivot, I don't know if you saw the website.

I did. That looks very interesting. Yeah, so now...

After talking with customers and doing research, end of last year, I found a customer, a machine shop, who was willing to try our device. So it was a pilot customer, paid pilot, very small amount. Oh, you got the testimonial?

Yes, so she's the testimonial, yes. So she agreed to put that on there. And so we deployed on 30 machines and she's using it. So it's really great. And then she converted to full-time.

start full time next month so now we are just signing the contract for the annual yeah so So she's a small machine shop, she's not at the biggest contract and you definitely can now sustain and grow like to a big company with just those machine shops but it's the beginning for me you know.

Getting that first revenue is the hardest. So the fact that you've got revenue so fast is a really big deal.

Congratulations. Thank you. Thank you. Yeah, I'm really excited. And with her, she's introducing me to other potential customers. She helped me schedule a demo with another machine shop much bigger than hers. They have more like 50 machines. She has more like 10. So it's good. And also, they had a pilot with Intrinsic where where they had the robot arm there from Trinity and Intrinsic was putting their software on it and she had me meet with them at her machine shop there and now I'm gonna have a conversation with Intrinsic.

So they really liked my device and the dashboard and they said, "Oh, this is something that is not in our turf, you know, so we could potentially partner." So I'm looking at that kind of partnership because From what I learned, the more I learned about this market and industries that is fragmented in the sense. Yeah. It's smaller manufacturers and there are big ones, but it's harder to get to the big ones when you're small.

So my strategy now is to, you know, you know get the small ones as many as I can but then partner with like the big guys like Intraseq because probably they can you know distribute better so if we have some sort of symbiosis here then maybe that could be a good strategy for me so that's where I am at so I want to get your perspective from it, maybe get some advice if you have anything to say about it.

I don't know. I think you're doing exactly the right things. You're actually getting customers, getting feedback. I'd love to know more about the technical solution. If you want to give me a quick overview on just your ideas about how that's working.

The initial solution was to have Raspberry Pi. in there so that's what was in the box and then we had modules from STMicro for the accelerometer the microphone is from TDK and also just modules from at the beginning but I had kind of like it was kind of expensive to have a fleet management platform for the For the Raspberry Pi I used but Lena OS yeah, and and If you have a thousand devices, it's fine because it scales so it goes down to like two three dollars per month But at this stage is too expensive for me.

It's like yeah, so I just decided okay First of all, even the box itself with the Raspberry Pi. It's more like 60 dollars to build now So I decided to make my own version or with a microcontroller with an ESP 32 So the price goes down to 20 bucks Also, I got to learn how to build my own management platform system on AWS. So I have the IOT service. So it was really nice to learn about that because I never got the chance.

So now I learned how to do that. And of course, with the co-pilots today, like, you know, I use it so fast. I mean, I just imagine it. I know how to do it and to code stuff, but it just does it for me. It's like I think it and they do it as I would do it almost.

Yeah, I mean, it's been going so far. Like I've had the same experience. has been incredibly accelerating in terms of just being able to put things together.

So that's great for me because Even if it's just myself now, so I hired actually a salesperson. who is the friend of my first customer. So she's part-time 20 hours per week. So let's see how it goes. This is her fourth week. So let's see. And then I have a friend who's helping me part-time as well as a consultant. But even with just a very, very small team, I feel like we can make technical progress very fast because the Yeah, and Tropic is helping you just go very fast.

Yeah, yeah. But I do, as I talk to... Some potential leads, you know, or even just talking here at Plug and Play with the partners that are in different verticals, I can see them being a bit like, you're kind of a small team. So I do wish to grow a bit, to have either not a big team, you know, three, four people, maybe five, you know, five in total. This way, I know that I have the AI multiplier, like the cloud multiplier there to make progress.

nice team there. Yeah. So, yeah. I think like nowadays, small, nice, compact teams can make a lot of progress. So, yeah. Absolutely. So yeah, now I have the ESP32. I just finished up a new, so even that was built with modules. We changed the accelerometer to the one from Bosch because we have a partnership now with them. So sign in to get one of the latest accelerometer with the highest frequency possible.

Because that's another thing we want to detect. We want to detect vibrations from the machines. And I do have now more than three months of data and I've been processing it. Yeah. Don't have enough time. I'm trying to give that project to my friend who's helping part-time. Yeah. But in those, In this data I was able to see the utilization of the machine. I didn't see harmonics and high frequency yet where you could detect that the machine is going to go down or something, but you could see utilization.

So you could tell people the capacity of your factory quite well. You can see spindle time basically for each machine. So it was pretty cool. And so what's the primary sort of information that you're giving to customers?

Yeah. So now I give them, All the notes that the operators take So like any like preventative maintenance tasks that they have to do and that's important for the factory owner because they that's a requirement for the certification ISO certifications or other type of certification and and sometimes they don't do it so she told me like I want to see that they do it. Yeah, so that's better.

They do that through voice?

Yeah, so that's the point. The main point is so that they do it through voice, so they don't have to take notes. And even if the notes are there, then she has to put them in the computer. So she told me she wastes time to do that. And this is just a simple solution to give them all the notes of what's going on. Eventually what I want to do is that as we get more notes and more notes and not just about preventative maintenance but also about like Weird things that happened to the machine or interesting solution that they came up with.

they can talk to the box, say it's there and it stays, it's stored. So as year passes, months passes, or somebody new comes in, they could ask to a little chat assistant inside the dashboard. How did we solve it last time? So kind of a tribal knowledge accumulation thing for the whole factory. So that's where I'm going long-term with.

Is there any, Do you see any demand for people being able to ask questions that would be answered by the manual? or things like that?

I want to go there too, yeah. Definitely. I haven't, like this particular customer didn't ask, but I think other ones will as it grows, yeah. Like different type of customers, like bigger factories, I think they will need that. And eventually I think all this solution, like I talk a lot with Bosch, they are They are the our partner, they're trying to connect us with other factories and companies. They have a lot of connections in Italy because the people I know are from Italy.

But yeah, It's interesting to know they say they keep claiming that a lot of them want everything on edge Which is true, okay, fine, but I... The customer I'm talking right now with, they don't care yet because I think they're too small. But eventually, the ones that require-maybe the machine shops that work with the government, they do, I don't know. the parts for like satellites and things like that.

They are more even I want to see one here. They were kind of like, oh, have the badge do the sign in all the security. So I think they would like an edge solution. So once you have the edge solution, then maybe we can use Moonshine, you know?

Yeah, yeah, and I just threw something into the chat. This isn't actually even using voice yet, and it's not necessarily... It doesn't have to run on the edge. It's just... I'm working with last with Pi. to basically create a help botOkay. has analyzed all of their documentation And then you can ask it, you know, questions about like, hey, why, what do the LEDs on boot mean? Or, Why isn't audio coming out of my headphone?

Oh, okay. Yeah.

And I thought it might be interesting because The technique I'm using, which is to scan documentation and then come up, have an LLM come up with It's called like hypothetical questions for each section. Like questions a user might find answered in that section. And then just doing a sentence embedding match between questions that somebody actually asks, and those hypothetical questions, it actually seems to be pretty effective and pretty generalizable.

Oh, that's interesting. Yeah, because this be used to load the machine, the CNC machine or whichever machine manual Yeah. And yeah, synthesize all those hypothetical questions and then have the box. If you remember, my box at the beginning had the speaker too. Now I limited the speaker, but I plan to have it eventually. So we can answer questions like this. That could be interesting to embed this kind of solution.

Yeah, because I was impressed by this approach. It's always... I've never had a good way to ingest arbitrary documentation. Like the stuff I have been doing... I don't know if I showed you the dishwasher demo. No. Okay, I'll throw that in the chat, and like 40 seconds long. Um, why am I dishingOkay.

Should I watch it now or?

Yeah, you might as well watch it now since it's short.

is still dirty.

Check if you used enough detergent and if the dishes were loaded properly with soiled sides facing down.

Why isn't my dishwasher draining?

If you notice water left at the bottom or if your dishes are wet after a wash, it may not be draining well. Check the drain air gap if you have one.

My dishwasher seems to be leaking.

Check the door seal for any cracks or dirt. Make sure the dishwasher is level 2, as an uneven position can cause leaks. If the problem continues, contact a professional.

What are you running this on?

So that's running actually on a Synaptics like $5 chip. And it's a commercial project we did for Panasonic under contract. The Uh... It was... the only thing that wasn't great about it was that we had to kind of put together the FAQs manually. Ah, okay. But other than that, it's using the sentence embedding lookup. So I'm pretty excited now I've found a way to create the FAQs Automatically and a lot of our customers, they want to make sure that the answers that are given are correct.

which means either using stuff directly from the manual or summarizing from the manual ahead of time And having people look through the answers to make sure there's nothing dangerous happening.

Interesting, interesting. So now you guys are mostly focusing on software only, just all the algorithms. Have you seen any need from your customers to have custom hardware? Or like they all make their own custom hardware.

So right now, um... Like the nice thing about the Panasonic thing was we were actually using the SOC that's, or similar SOC to the one that's already in there. Um... their appliances. So anything where we can, because having custom hardware basically adds like a couple of years To the... deployment, unless it's something like what you're doing where it's kind of like a box you stick on the side. Yeah, yeah.

What the f*** is this? Yeah, I'm still... I'm trying to debate with myself if hardware makes sense, but I want to give a simple solution where you can extract this knowledge that is lost from factories. So that's my goal. You know, sensor and IoT part, particularly maintenance, is more like, so I also offer that because there are some companies that already do that. Or at least they claim they can do that.

Sometimes, I'm not sure if it works well, but at least you have that as a base. And then you have this new thing.

And my experience with predictive maintenance, because it's been something that has been coming up you know, for a long time, at least like 15 years, is it tends to be very bespoke. So you end up doing... something that will work with one model of machine. And then you kind of have to do a lot of work for the next machine and it's very hard to generalize. Yeah, yeah. So I just, I've seen Because there's obviously customer demand for that.

So I've seen multiple startups kind of go down that route and it just turns out to be something that isn't Scalable.

Yeah, if you can generalize, you can have scales.

But what you're doing sounds great because, you know, that is way more generalizable and it sounds like it ties into some, like, regulatory...

requirements as well which is always a little bit something because that's like taxes like taxes are for everyone regulation for everyone yeah um yeah i'm okay to do you know for the beginning a little bit of customer just to get my door so i even get experience and all that some trust from customer that can present introduce me left and right but yeah always i'm keeping an eye to make sure i don't go down a rabbit hole where oh this is the onlyYeah, and you're not like, You've got a lot more than predictive maintenance that you're offering and all that stuff is, you know, totally replicable across multiple customers.

I think you're in a great spot. Fingers crossed. I feel like I'm getting a bit more traction, especially just putting the website out there. It helps a lot just having conversation with people. I feel like I underestimated the importance of having a website. I had the lore of the landing page, you're mysterious, tell startup. I feel like it doesn't serve any good to anyone. That's great if you already have people you yeah know who you are because they're not actually going to your website but if not it's yeah yeah maybe a second third startup once somebody asks me yeah So yeah, now what I'm trying to do in the next few months is generate a bit of more traction with customers for sure.

So I have a few presentations here at I plug and play with some of their industrial partners, a few workshops here in the Bay Area with other machine shops. So I'm trying to do that. And in parallel, I need to raise a seed round, basically. So how would you go about, like, how do you go about raising a priestly drama? How was your experience like after the angel run?

Yeah, so... It's...

Like, if you can find, um... People who know the problem space really well and maybe have a strategic interest. Um... So, for example, you know, Bosch has a venture arm. And this might be, you know, kind of way too early for them.

Yeah, yeah. I talked to them. I have a friend, actually one of my neighbors. She's in Bosch Ventures and also my advisor and the other investor, just an angel like you. He's also there, but the Bosch Ventures is more for Series A. Yeah, exactly.

That's going to be... Ah. Really, I mean, I think it's in my experience, it's very similar to the angel round. You kind of have to get people who you have a connection with and people who've seen the results. Like you're probably not going to be looking at institutional...

Yeah, I would try to stay away from institutional if possible for this round because I just imagined this to be... So Adrian round was 100k, I imagined this to be 500k. Yeah. Yeah. So... Even I wouldn't even raise The valuation, the cap, I would even probably keep the same. What would you suggest there? Um...

It's... Honestly, it's kind of a tricky spot. I would think about... Um... Like one strategy is to just kind of keep extending the angel round. Okay. So instead of trying to get 500K in one go, Like, unless you have... Um... Like... you know, somebody who specializes in those kind of pre-seed rounds already lined up I think your best bet is going to be trying to find more angels as you're doing your you know, as you're doing your, uh, customer discovery and kind of market discovery.

Basically, and I would also talk to like your connections or ventures and say, hey, you know, who are, who are well known angels in this area That's an interesting point, yeah. I've, uh, I haven't done it myself, but I have, you know, I've heard decent things about AngelList.

Angel, okay, okay. Multiple times this name came up, okay.

Yeah, I mean, just because it's the main kind of public way of finding angels, angel investors. Now, there are going to be a bunch of downsides, especially because you're going to have people who are Um... Not necessarily that sophisticated. You know, the classic profile is like a dentist.

Yeah.

And the problem there is that that they're not actually used to risk investments. You know, like they're used to like a dentist, you basically rent an office, put a sign out and you start getting money coming in. Yeah, that's not. Yeah, exactly. And this this is pretty different. So you can end up with. investors who are hard to please. I see. But, you know, as long as you're making sure you qualify them so that they you know you make it clear to them what the what the deal is Because...

Like trying to go out and raise $500,000 in a pre-seed, You know, that's... it's not likely to come from one or two people. Yeah. because it is pre-seed and you probably aren't going to get institutional so then it becomes much more of a role you know think about it much more as a rolling hey can you get 50k from this person and you know 25k from this you know this person and just kind of keep raising um as As you go.

Okay, that's interesting. And what are your thoughts on those accelerators? I'm trying to make a list. I've tried to apply it a few. Actually, in January, I had a YC interview. Oh, yeah.

I'm not sure if I told you. Yeah.

but Their main concern was that first, my team is small and I don't have a co-founder. Yes. And that this thing won't scale fast enough or big enough in the three months of batch. And Rich was like, yeah, okay. Of course, but it's not software.

Yeah, no, exactly. It is tough generally for anything that's physical because they are going to look at that and as soon as you're shipping stuff and having to get stuff set up in locations, then that's a lot of overhead.

It's interesting. I'll argue even into this one, if you look at Nvidia, solo founder, hardware.

Oh no! I mean, that's the thing that drives me crazy is Apple. One of the most valuable companies in the world, you know, the whole thing is built around hardware. Yeah, and it's ironic Yeah, I've got a whole rant about how Silicon Valley does doesn't actually invest in any kind of silicon anymore.

I think it's going to be a comeback with the AI chips and stuff, but it's very specialized and the people that raise hundreds of millions on that Yeah. I don't know. They have some good connections. It's not going to be me now. Maybe next time. So yeah. So I remember you introduced me back in the day to James from Bloomberg Beta. Oh yeah. Like he passed like on the opportunity back then and was still the medical device back then, you know?

Yes. And... I think it's worth reaching out to him again.

I'll definitely reach out to him again.

It's like institutional money, so probably still, but it's good to keep in touch.

So I'll definitely do that.

So I wanted to ask if you have other, maybe someone who could introduce me like more like even earlier stage that could lead to any good conversation. Yeah. So send me like two or three, two or three sentences that I can forward. And I will, I will figure out some people to kind of circulate it.

Okay, yeah, that would be great. And now there is a website they can also look at, this, like all these people, person is serious. Okay, yeah, let me do that. Adio. Yeah, and one last thing. I don't know if this is the best place to stay long term at Plug 'n Play. My contract for the desk ends in June. Do you know any good place in Mountain View? Is the building where you guys are affordable?

Yeah, no, we found it pretty good. And I believe that you can... I'm trying to remember if you can... But you should be able to enter desk, I think. Oh, here it's spotty too, so... Silicon Valley!

Sometimes you do, but you need to set up the meeting and everything so it's almost like not being here. So I'm looking for a place that is a bit more vibrant and it's easier to connect. Because that's what I especially need now at the beginning. Yeah.

Awesome, this is fantastic progress and I look forward to your sentences.

Great work. Thank you, thank you. I'm looking forward to see more demos from you guys from Moonshine. Hopefully in the future when we have to implement some edge solution maybe we can use your model and we can win-win here. Hopefully in the future when we have to implement some edge solution maybe we can use your model and we can win-win here. Awesome. Okay. Thank you. Have a good weekend. Bye.