What comes to mind when reading the term “no-touch serialization?” and how does that fit into the machine vision space?

Can you imagine taking images of each blank piece of paper in a full ream, “feature printing” (think digital fingerprint), each of them, and then being able to identify each one? For U.S. Pacific Northwest company Alitheon, this is everyday life. In this episode of Manufacturing Matters, Roei Ganzarski, CEO of Alitheon joined TECH B2B Marketing’s Jimmy Carroll and Dan McCarthy to discuss the technology and how it’s used in industries including transportation, luxury goods, collectibles, art, government, and beyond. Ganzarski also talks about the differentiation between barcodes and feature prints, how AI comes into play, how customer data is handled, and the intersection of this technology and machine vision.

Jimmy Carroll: [00:00:07] Hi everybody. My name is Jimmy Carroll. I’m the vice president of operations at Tech B2B Marketing. And welcome to this episode of the Manufacturing Matters podcast. Today I have the pleasure of being joined by Roei Ganzarski and my colleague Dan. Roei, thanks so much for taking the time today. I really appreciate it.

 

Roei Ganzarski: [00:00:21] My pleasure. Thank you.

 

Jimmy Carroll: [00:00:23] Yeah, of course. So let’s start off by just letting everyone out there know, like for people who aren’t familiar with the Alitheon, can you tell us a little bit about the company and what you do there.

 

Roei Ganzarski: [00:00:33] Yeah, sure. Alitheon, we’re based in Bellevue, Washington. So the great Pacific Northwest of the U.S., and we are a no-touch serialization company. Basically, we serialize things just by taking their picture. If you think of the most equivalent thing that we all are aware of, we fingerprint things just like governments would fingerprint people. That’s what we do.

 

Jimmy Carroll: [00:00:53] What are some of the primary industries you serve?

 

Roei Ganzarski: [00:00:58] Yeah. So, to serialize products or items, we serve mainly five key industries. One is luxury goods, collectibles, art –basically expensive items that you don’t want to get wrong. The second is transportation, so car parts, plane parts, etc. Third is health care, so things that go in and on your body. Fourth is precious metals, the type of big gold bars that you see in the movies going into banks, etc. And then finally we do government work, Air Force, Army, etc. Basically we define it as high-consequence items. If you don’t trace them correctly or you get hit by counterfeits or the gray market, there is a high consequence, either financial or even safety and risk to life.

 

Jimmy Carroll: [00:01:40] Okay, so this is fun. Dan, I don’t want to go too deep, but I’ll ask one more question before before you chime in if that’s all right. Roei, so for manufacturers — the name of the podcast is Manufacturing Matters — but for manufacturing and other businesses out there, beyond just the manufacturing plant floor, those that are leveraging industrial automation technologies today, what are some of the biggest pain points that these companies are facing? Obviously labor shortage, but what else?

 

Roei Ganzarski: [00:02:11] Yeah, I mean, there’s a few things. If you think about in-plant, there’s a lot of talk to get to that holy grail as they call it of Industry 4.0 or Manufacturing 4.0. Basically, a lights-out factory, that the data and computers are running the factory in the most efficient way. Right now all of the focus is on the machinery. My machine tells me through an app what’s going on with it. How many parts went through it? Is it in sync or not? Is it operating? So I get this great overview anytime of day about my factory. What a lot don’t have is: What’s going through that machine? Which parts, which items were made by which machine at what time, and why is that important? Let’s say one of the things of this Industry 4.0 is being able to adjust scale, adjust tolerances, adjust accuracy in real time, right? Let’s not wait until something goes out of tolerance in order to fix the machine. But what that means is I need to know what’s going through that machine in order to do that. Imagine I’m now at the end of the line, and I see my products and through some automated quality control, I know that everything is within tolerance, but I’m starting to see that some of the parts are starting to edge towards the upper edge of the tolerance, which means that at some point, they’re going to go over it and I’ll have a flaw.

 

Roei Ganzarski: [00:03:28] Well, instead of just trying to guess which parts went through which machine, imagine that with full implant traceability, I could know that those parts that are all edging towards the higher end of the tolerance, they all have one thing in common. They all went through machine no. 7 in the second night shift. Okay, great. Now I know where to go look and what machine to deal with. So being able to do traceability of parts level or items level at the plant, really, really important and right now is lacking, because a lot of parts can’t be marked, they can’t be somehow identified. And so being able to do it with no-touch serialization, which means just relying on a picture, is really powerful. The second problem is, once it leaves my plant, where is it? Who’s the last person that saw it? Did it really get stuck somewhere in the supply chain? And even worse, is someone trying to counterfeit my products? And will my customer come back to me and say, hey, your computer board, your watch, your disk, whatever it is, your brake pad, failed. And will I find out: Wait a minute. That’s not even mine. Someone faked it, but my brand is on it, so now I’m responsible. So those types of things all boil down to the same thing. Can I irrefutably identify my individual item and trace it and show provenance back to birth?

 

Dan McCarthy: [00:04:45] That was a very succinct and yet comprehensive response, Roei. But I think you’re going to kind of cover some of the same ground with this question. But companies already utilize data codes, lock codes, etc. and 1D, 2D barcodes, fairly inexpensive solution as part of various standards. So what’s the unique proposition for Alitheon? And do you see your devices as complementary, competitive with barcode scanners? I mean, how do they fit in with the larger scheme here?

 

Roei Ganzarski: [00:05:14] Yeah. You know, that’s a really good question. Proxies, as we like to call them, are really prevalent. Etchings of serial numbers, to your point, barcodes, QR codes, data matrix, hologram stickers. You know, all these types of proxies all exist today, and they are fairly cheap. They’re also very problematic. And what do I mean by that? I equate it kind of to ourselves, if you will, to fingerprints for people in the beginning. What we do is we fingerprint, or we call it FeaturePrint, things. Why do governments, airports when we travel etc. want our fingerprints? Because the proxy we carry for people, it might be a passport, a driver’s license, a badge. The same type of proxy can be lost, forgotten, manipulated, handed to someone else, or even itself counterfeited. So when I go to the airport, for example, I give my passport and still I put my hand on the fingerprint reader and they want my fingerprint. Why? I gave them my proxy. The passport is telling them who I am. They want to know, is it really my passport or is there some manipulation going on? Maybe we have teenage kids that have fake driver’s licenses that allow them to say, Hey, I’m 21 and I can go to this bar and drink.

 

Roei Ganzarski: [00:06:19] Never happened, of course. You know, no, no, no. But maybe. That’s the same thing. If you trust the proxy, you have a problem. Imagine twin sisters, one wanting to cheat on a test for her twin who didn’t prepare. She takes the driver’s license, walks into the classroom, and says, “I’m Julie.” But Susan is the one taking the test. But imagine if fingerprints were tested. Even identical twins or identical triplets have different fingerprints. You wouldn’t be able to cheat. And so the whole idea behind what we’re doing for products or physical items is: Let the item speak for itself and be its own identifier. While barcodes and QR codes and etchings and stickers are nice, at best, they’re identifying themselves. I am a real QR code or a real data matrix or a real hologram sticker. I may not necessarily be on the real or the right product. By the way, it may be unintentional. Wrong stickers, wrong products, wrong QR codes put on real products but then create confusion. And so we’re basically saying: Don’t rely on a proxy or an additive to tell you who the product is. For example, when its expiration date is. Imagine you’re a manufacturer of medical care or pharmaceuticals or medical implants. Do you really want to trust the expiration date on something printed on the box? Medical pharma has over $400 billion of fake and fraudulent goods worldwide.

 

Roei Ganzarski: [00:07:46] How is that done? I take the box. I erase the, for example, expiration date, and I print the new one on. And then I get the medicine as a consumer, I look at the expiration date, I’m happy, except the medicine is already expired. Now maybe it’s one day expired and no big deal. Maybe it’s three months expired. Who knows what that could do. Creams, parts for aircraft, etc., so the idea that we’re bringing forward and delivering to our customers is: Rely on the part itself that you have feature-printed at birth, meaning at manufacturing, and not on any proxy that makes it convenient. Now for some I will say we are a complement. Why? Maybe some want that QR code to open up this cool website. Having a customer experience. That’s great, right? Some labels give you the required legal elements of where was this made and so on. And you don’t want to complicate it, or that’s an easy solution. That’s fantastic. But we can be on the side, protecting the actual product while the label just gives you some sort of information.

 

Dan McCarthy: [00:08:47] Wow. Yep. Okay. It’s a solid answer. And a little bit of a devil’s advocate question, but your solution is software-based, correct? You’re relying on the imaging technology available to the user as well as the user’s understanding and ability to capture high-quality images, lighting, all sorts of factors that they may not be taking into consideration for optimal results. How do you ensure accurate results with your technology?

 

Roei Ganzarski: [00:09:12] Yeah. You know, that’s a really good question. The same questions came up when, if you remember, in the U.S. at least, and now in other places in the world, we started to deposit checks through our phones as opposed to walk into a bank. And those same questions were asked. How do we know that they’re going to take a good picture of the check? Well, in the beginning it was a little difficult. But today it’s all natural. You open up the app, it shows you a little square there. It says, put the check in the square, push the button, and if it’s out of focus, it’ll tell you: “You’re out of focus. Do it again.” And by the way, check capturing is really easy because you’re really looking for that set of numbers on the bottom of the check. What we do is fairly simple and very similar as well. First of all, we worked really hard to get to a point that we use standard off-the-shelf cameras. We don’t offer or require any proprietary cameras, spectral imaging, X-ray. Nope. Take a standard, in the case of manufacturing, industrial camera like those made by Basler, Allied, Cognex, it doesn’t really matter. Install it on your production line, and that’s the only thing that goes on your production line. There’s no new machinery. There’s no new process. There’s no touching of your part. You just install a camera with a little arm on top of it, and there’s some sort of optical trigger maybe or mechanical trigger, however your line works, and every time a product goes by, a picture is taken and our software will take that picture and convert it to a digital fingerprint or what we call a FeaturePrint.

 

Roei Ganzarski: [00:10:29] So that’s on the easy side. Standard, lights-out, APIs will take the data. No people involved. At the other side of it, maybe you enable one of your customers or your QA person or your consumer to pull out their cell phone, because we can also do it with mobile phones, snap a picture of that and also be able to identify it and say, yes, this is a real phone, a real Samsung phone in this case, that I just bought on eBay, as opposed to a fake one that I bought on the corner street in New York. Maybe. In that case, it’s also very simple. You have, just like with your check deposit, you could have a little crop or an overlay of the image that says, align this thing on your screen to that product, and the picture can be taken. Lighting is also very simple. The beauty of phones, they all have torches. And so what we can do is make sure with our app that the torch is on, for example, and that overcomes any ambient light change. So maybe you’re in a well-lit warehouse, maybe you’re outdoors and it’s a little dark. The torch overcomes all of that. So the beauty, and that’s why some of the power of what we’re able to bring to market is, we’ve gotten to this point where computers and industrial cameras are very affordable and very good. Phones and their cameras are very affordable, very good, and very prevalent, and our algorithms come into play. And between that kind of convergence of those things, we can now do what people have only dreamt of doing, which is take a picture of an item and serialize it to the item level.

 

Jimmy Carroll: [00:11:56] So this is all really interesting. So when I think of a camera on a production line and a typical machine vision system, what I’m picturing or what people consider a machine vision system is a set of components: camera, lighting, lens, software that ultimately makes a decision, usually in fairly high speed. Immediately, right? But yours is sort of on the other end of it. So you’re capturing images and storing these images in, and correct me if I’m wrong, this is sort of walking through it. You’re storing these digital FeaturePrints, right? And you’re storing these in the cloud, I imagine somewhere in a repository for individual companies and part types and whatever different data silos. So how does it work? So once you capture the images, so you have images of a part or whatever it may be, and using this example as the industrial space, later somebody wants to go verify that a PCB was theirs and not somebody else’s. So how does the AI work at this point?

 

Roei Ganzarski: [00:12:57] Yeah, really good question. And I’ll start with an analogy, back to people’s fingerprints. If you ever watch “Bourne Identity” or “CSI” or any one of those police movies and they find a fingerprint and they’re doing a match, and sometimes they show the screen and you can see kind of this system running through the fingerprint and looking for various dots. There’s no picture of people there. You don’t need the picture of the finger. What you’re looking for is the data. When we take a picture of the item, let’s say on the production line, as that picture is taken, our software, more importantly our algorithms, immediately, on the edge, take the data from that image and create the digital fingerprint or what we call the FeaturePrint file. By the way, it’s smaller than 400 kilobytes, tiny. We don’t need the image at that point. In fact, for most customers, we don’t keep the image, because I don’t need it. One, it’s big. It’s expensive to store. It’s expensive to move. Two, it gives you the ability to theoretically reverse-engineer. Oh, I have a digital fingerprint and now I have the image of it. Oh, I could kind of look at what you’re trying to do. Now you couldn’t. We have a really good set of algorithms, but let’s say, for security purposes. And then there’s that element of and now I’m going to put all the data related to it as well. I don’t want your images.

 

Roei Ganzarski: [00:14:10] I don’t want your data. I don’t want to get involved with all those things that may be of risk. All I have, if you were to somehow break into the security of the cloud and break into our own security and obfuscation and encryption and all of that, and you got to our database and saw digital fingerprints, or FeaturePrints, what you see are files with a bunch of 0s and 1s. There’s no graphical element to them. There’s no data to them. You’re basically seeing digital fingerprints with UUIDs, basically serialization of our digital fingerprints. So let’s say you opened one up and saw a bunch of 0s and 1s. You wouldn’t know: Is that a gold bar? We protect gold bars. Is that a watch? We protect watches. Is that a brake pad? We protect brake pads. You wouldn’t know. It isn’t written down anywhere. There’s no image anywhere. All it is is a set of data points that say: This is what this digital fingerprint constitutes. Now, go into the future, someone pulls out their phone, takes a picture of that item, using the crop and the overlay and they took a good picture. Now our system on the edge again creates a new digital fingerprint, or FeaturePrint, and sends that back to our system. And now that match happens or the search happens. Are there any data points and digital fingerprints that match this new one I made? And our algorithms, let’s say they find one, might say, “Yes, this is it. This is item no. 176532-7. And now I send it back to my customer system, who then says, Oh, that is a brake pad manufactured at this state on this machine, etc., etc. So it’s in fact as simple as a trigger, if you will, to say you’ve just, or someone has just identified an authentic item that you made. But what that item is, I don’t know and I don’t need to know. And that’s the beauty of our system, as opposed to the traditional AI or what we think about today as traditional AI, which is machine learning–based. I have to train my system, so I have to know what it is, and any variation in the item I have to retrain it. We don’t train the system at all. We just start taking pictures of items because I don’t need to know what the item is. Our algorithms are completely item-agnostic. I don’t need to know if it’s a stainless steel back of a watch or a composite brake pad or a titanium part of an airplane or a canvas of a piece of art. It doesn’t matter. Our algorithms know how to search for things that make it special or unique, and they codify that into a digital fingerprint. So not only is it really easy from an identification perspective; it’s also very secure because I don’t have or need any data or imagery.

 

Jimmy Carroll: [00:16:42] So how does it work on the other side. So, maybe not a good example, but I’m just thinking of something that would have really bad ramifications.

 

Roei Ganzarski: [00:16:55] Brake pads on a car.

 

Jimmy Carroll: [00:16:56] Well, brake pads, brake pads would be a perfect one. I was thinking like an EV battery.

 

Roei Ganzarski: [00:17:01] EV battery, yeah, same.

 

Jimmy Carroll: [00:17:02] And the part’s not sealed and it explodes. And maybe “explode” is severe, but you know where I’m going with this. Somebody wants to bring that back to you and say, “Is this your client’s part?” How does it work then? You know, in terms of looking at all the 0s and 1s and lack of images. How does that work?

 

Roei Ganzarski: [00:17:25] So go back to fingerprinting. If you were to cut off your fingers, I can’t really help you. There’s no fingerprints to be found. But if you have scratches on your fingers or even a partial fingerprint – you hear that a lot on TV: “Oh, I have a partial” – they can still identify you. We’re the same thing. If I had, let’s say, to your point, a battery, and that battery in fact exploded and there’s nothing left, I can’t help you. I mean, at best, the data can say I installed or the installer put in that battery, but I don’t know. But what we’re here to do is preempt that. What do I mean by that? Where could batteries be problematic? One, I either have a fake battery coming into my car. That’s one option. And by the way, while people think that’s remotely not possible, in Europe they’re already finding CATL and other battery manufacturers with fake UL and certification stickers on them. So this is already happening, right? The more there’s demand for something, counterfeits and gray market will grow. So there’s already fake batteries. The other option is, someone had let’s say a total loss in a car. And I’ll just make it up.

 

Roei Ganzarski: [00:18:25] It was a Tesla. And Tesla said to the shop, Hey, that’s a total loss. They’ll get rid of the batteries or return them to us or damage them, but don’t use them because you don’t want to reuse a set of batteries that went through an accident. And maybe that company says, Oh, you know what? I could make an extra dollar here. I’m going to sell those batteries to some other company because, hey, they look okay. In both scenarios, imagine you’re now a new installer. You just bought batteries that either you thought were new or you think are used but good condition, meaning allowable recycle. You could pull out your phone, take a picture of that battery cell down to the battery cell, and it could tell you either: I don’t know what this is. So it’s not a real battery, meaning an authentic one, or, oh yeah, this is battery no. 17653-9 that was made and installed 12 months ago. So if you have paperwork that’s telling you it’s new: red flag. So that’s how it would work there. But if the battery’s exploded and there’s nothing left, it’s just like I would find you without any fingers, I can’t fingerprint you.

 

Jimmy Carroll: [00:19:27] Yeah, yeah. That’s why I stopped and said that’s maybe a little severe, but yeah, you knew exactly.

 

Roei Ganzarski: [00:19:31] Wear and tear. I can tell you, for example, anecdotally we work with some luxury goods companies, and they did some damage testing, as we do with some defense contractors who did damage testing. And a lot of them went through, anecdotally, I can share not who but what: Some of them took metal grinders and put them against the titanium part, and according to their internal tests, damaged 90% of the face of the item and they were still able to identify them. Others have taken parts and took them through 25 years of what they call accelerated damage tests. So they put them in these deburring machines, etc., and they create simulated damage over 25 years. Still able to identify them irrefutably. And so it’s fairly robust to wear and tear, damage, UV, etc., until a point where there’s nothing left.

 

Jimmy Carroll: [00:20:20] Roei, that’s really fascinating. I want to ask about some general trends in technologies in the space now. AI is obviously a topic that’s at the forefront of a lot of people’s minds, both industrial and in the general public. It means a lot of different things to a lot of different people. In terms of manufacturing and the spaces that you serve, beyond what you guys are doing, what are some ways that you’re seeing AI benefit businesses today?

 

Roei Ganzarski: [00:21:07] Yeah, so I’ll touch on a few things. But there’s also one more new technology or maybe it’s not so new anymore, but that’s used a lot is blockchain. So I hear a lot about AI and blockchain. And I’ll touch on both of those. We would be considered a branch of AI in that we are machine vision–based. And machine vision is considered a block of AI. We’re on the deterministic side of AI, which means we actually give you with set algorithms a set answer; versus the generative AI that we all are excited about today: ChatGPT, Copilot, etc., that’s more on the stochastic side of AI, which is, I’m going to take a bunch of data. I will learn from it and then create something for you that may or may not be good, may or may not be suiting. And it’s at a percentage base probably okay. What do I mean by that? Take the issue of counterfeits. So I’m a manufacturer, I’m releasing products to my customers. Let’s say I’m even manufacturing them for the military. I want to make sure that my customer is getting my real products. Let’s say these are computer boards that go onto an aircraft or a submarine. I want to make sure they’re authentic and that they don’t have something with my brand on it that’s not authentic. There’s two ways you could think about that.

 

Roei Ganzarski: [00:22:21] One is, let’s train a machine learning–based AI. Let’s show them 10,000 real computer boards. And let’s show them 10,000 fake computer boards. And then similar to how a quality control system would work, you would take this to the next level. And the military, let’s say, would get a board, take a picture of it, and say, Yep, I think that’s a real board, 90% chance that’s a real board. Great. That works nicely. What it won’t tell them is: By the way, that board is a used board that was already installed two years ago. So it’s actually been harvested for something old. It’s not what the new manufacturer just sold you. But it’s real. So the challenge with this type of learning-based system, deep learning, machine learning, etc., is that you’re solving one of four problems that exist in industry today. The problem that it’s solving is, for the most part, stochastically, counterfeiting. It doesn’t solve gray market – recycle, refurbished, etc. –stolen. It doesn’t solve traceability and it doesn’t solve human error. Why? Because those require the ability to identify the individual item. If I want to know the provenance of one item, I need to trace that one item. I equate that to in a crime scene, going back to fingerprints, I’m a police officer, I find a fingerprint, and the AI learning system says, “Oh, that’s a human being that committed the crime.” Well that’s interesting that it wasn’t a wolf or a shark, but you really want to know which human being? Serialization.

 

Roei Ganzarski: [00:23:52] And that’s what fingerprints give us. Digital fingerprints, or FeaturePrints, same thing. What we provide you is not: That is a real part. We’re telling you that’s part no. 1753-2. Now if I know that, I can trace that part. I know that it’s authentic. And I know that you’re about to use the correct part. So I’ve solved all those problems. So that’s one element where AI can be used. The second is, AI can be used in QA. So I mentioned at the beginning, I talked about an example, let’s say you get a bunch of parts and the human being checks them and says, yeah, it’s still all in tolerance, but I’m seeing a trend of going into the right side or the high side of the tolerance, and AI might be able to catch that way before a human being does, because it could look at all items instantaneously and say, you’re not only within the tolerance, you’re within 5% of nominal and that’s great. But I already see you cropping up, and I see these 10 items every time cropping up. Now complement that with our system that will say, “Those 10 items all went through machine no. 3 on shift no. 2, when Jimmy was managing the machine.” Ah, well, let’s talk to Jimmy and see, “What were you doing during that time that put those all to end of tolerance?”

 

Roei Ganzarski: [00:25:06] But the combination of that type of AI that would look at trends and data, and then ours that would then take that specific data and point it back to a specific machine, that’s where that power comes in. So that’s an example of where the machine learning–based AI can work on looking for trends, looking for data, looking for things that can improve efficiencies, for example, making sure that something has been installed. So let’s take an aerospace company that forgot to put some plugs in a door or some bolts on a door, and the door falls off midair. That would be problematic. And AI might be able to be taught that before any plane is released, take a picture of that door, and it’ll say, I see four plugs in there or four bolts in there. You’re good to go. Or, I don’t see four bolts. Hey, don’t release the plane. Something’s wrong. So AI can help you on that. What that won’t help, where ours would complement is: Are these the right bolts? Meaning are they the ones that are made out of titanium versus the ones that are not coded for high temperature? I’m making it up, right? So, again, identify versus overall. But both those systems should complement. The second thing you hear a lot about is blockchain, and a lot of people fall into the trap of believing that blockchain can protect their supply chain or their manufacturing or their parts, etc. And that’s the best- and worst-kept secret in terms of the fallacy of blockchain. Blockchain is a phenomenal tool. We sometimes use it ourselves. It’s a phenomenal tool to protect data integrity. Nothing to do with the physical world, physical parts. So let’s go back to my Samsung phone. I may have data about the manufacturing of the Samsung phone and the supply chain of the phone that’s protected in the blockchain, which means I can make sure no one’s manipulating this data. No one’s touching the data. No one’s doing anything. But it doesn’t tell me if it belongs to this phone. It just tells me the data hasn’t been manipulated. And the way people trick the system today is: I’m going to take the same barcode that’s on this phone, and I’m going to put it and print it on the other phone. Now I’ve created over-trust. Oh, look, the data’s in the blockchain. That must be the real phone, because it has the barcode that took me to that blockchain. It’s like, no, it’s worse. Now you trust the data and you think it’s related to the physical world. So that’s again where you want the complementary aspects of; We’re going to tell you this is in fact phone no. 12345. And now let’s open the blockchain that says, “And here’s the protected data for that phone.”

 

Jimmy Carroll: [00:27:42] Yeah. Well correct me if I’m wrong here, but it’s kind of seeming to me like your technology, back to Dan’s earlier question, could be very, very complementary to traditional machine vision. It’s almost like an intersection of, in terms of AI, in the industrial space, two of the applications that get brought up quite a bit are visual inspection and predictive maintenance, predictive analytics, and it’s almost like using machine vision in your technology. It’s like an intersection of that, where the traditional machine vision system might be looking at these parts and saying, yes, these these passed, but your system might be looking at it saying, well, yes, it passes, but these are acceptable variations and these acceptable variations are growing over time. And soon this could create a problem like a door coming off.

 

Roei Ganzarski: [00:28:41] Right. Exactly. So that’s one aspect of it. Kind of like biometrics. We’re getting to a point where we could walk through the airport or go to an Amazon store, put our hand down, and I don’t even need to show a proxy. It’s just reading my fingerprints or my face, whatever it is, and I can go. But there’s also other sides of it, which is, I want to show my passport, and they want my fingerprints to validate the proxy. So there’s use cases for both, either as a complement to a proxy or standalone. But now let me put a twist onto the complete other side of manufacturing. Let’s say you have preventive maintenance on your machine and your AI or machine learning or the machine is saying, Hey, I’m about to have a problem here. You really want to change – again, I’ll make it as the simplest example, stupidest example – you need to change your oil filter before it breaks down. Great. Now I know to change my oil filter. How do you know you’re putting in an authentic, legal oil filter? What if you’re putting in a counterfeit oil filter that will start leaking a day into it. Now, yes, later your AI might catch it, but the damage has been created, so you could look at it on the other side, in your factory, forget the parts that are being made, where we already talked about, where technology like ours, non-touch serialization, can help. Imagine the parts and the tooling that go into your machinery. How do you know those are authentic? The tip of the welding machine that you’re about to use. Is that a real tip that will allow you to do what you’re supposed to do and have the warranties behind it or not? Is it a real filter? Go to the most basic, a real print cartridge. Now, if you’re the one that is going to buy a fake printer cartridge knowingly because you want to do that, okay, that’s knowingly. But if you think you’re buying a real one and you’re not because someone duped you, you could damage your printer. Now you maybe it’s a printer at home, you don’t care. Maybe it’s a high-scale, large-scale printer, industrial, or a 3D manufacturing aspect that could really harm your business. So the parts and elements going into your machinery, into your factory, you want to make sure are taken care of as well.

 

Jimmy Carroll: [00:30:52] Yeah. This is really interesting. One question from earlier that I wanted to ask you, and I think maybe you partially covered it, but I want to make sure I understand just out of my own curiosity. You mentioned that you just have to put a camera on a line, for example, or whatever the other applications may be. Do customers have to do any of their own AI model training? How simple is it for somebody to use your technology out of the box?

 

Roei Ganzarski: [00:31:22] It is as simple as you can think. As I mentioned, we don’t do training. There’s no AI modeling. Customer doesn’t have to do anything. All they have to do — and we just finished an installation not long ago — all you have to do is have a camera installed on top of your line. Let’s say you’re a PCBA line or a brake pad. It doesn’t matter. You look at where — and we can help with that — where are your products going in at the point that you want to start registering them? For some, it’s at the beginning of the line. For example, we work with one company. They register, what we call FeaturePrint, bare PCBA boards, and then as they go through their various pick-and-place machines, they will keep adding data, but identifying the bare until it’s a finished one. And they can still identify even though they’ve added a bunch of stuff on it. So they want to be able to trace it in-factory. Some might say, I only want to start tracing after the product is finished, so it really doesn’t matter, or maybe from the middle.

 

Roei Ganzarski: [00:32:16] Wherever that point is, you install a camera, and you install it on top, on the side. Again, depending on what it is that you want to FeaturePrint, or register, and you have a trigger that tells you — it could be optical, mechanical, software-based, it doesn’t matter — that says: A part is now going in front of the camera. Picture taken, automatically. That’s it. You don’t have to train the system, teach the system. We don’t even need to know what is the item that you’re doing. We work with some defense companies. We don’t even know what they’re FeaturePrinting. We don’t care. The system from that perspective is product agnostic. Or you could say so smart or so dumb, it doesn’t need to know what it’s taking a picture of. The algorithms are able to look at that image and collect the data that they need to create what constitutes a digital fingerprint, or FeaturePrint. So it’s really, really simple from that perspective. Users, customers don’t have to know anything about how or what we do; nor do they have to train for it.

 

Jimmy Carroll: [00:33:13] Yeah, I mean, that’s really interesting, because it’s sort of an obvious talking point. But over the last couple of years, especially within the context of AI and industrial, “ease of use” is a term that gets put out there a lot. Well, yeah, obviously everyone wants their products to be easier to use, but now there are a lot of new products coming out, like sort of fully integrated systems or software packages like yours that are just easy to use.

 

Roei Ganzarski: [00:33:39] Yeah. And the nice thing is, with some of the manufacturers we work with, they already have, to your point from before, machine vision systems on them for quality control. We might be able to use those same cameras. Again, we don’t make cameras. We’re a software company. We use traditional off-the-shelf cameras, even black-and-white ones. We don’t even need the color because we don’t care about color as part of an identifier. So we might be able to tap into the image you’re already taking for quality control and just extract the digital fingerprint, or FeaturePrint, out of that. So there may not even be a camera to install, depending on what you have there.

 

Jimmy Carroll: [00:34:12] Yeah, that’s nice, right, because something that comes up a lot now is, when people are investing in industrial automation technologies, they’re thinking of flexibility. Could I use this for another application later? Well, if you have a camera that’s already there, you can use it for something right now. Two things at once.

 

Roei Ganzarski: [00:34:30] Exactly. Or think of the manufacturers of the machinery for production that have quality-control cameras in them. Imagine if they were to contact us and say, “Hey, we want to introduce your technology as part of the machine, built in.” And we might tell them, “Oh, you have a 3 megapixel camera on that. If you just upgraded that for another 100 bucks to this other camera, you could put our software in there without any additions,” as a thought process, right? So depending on what’s already in place, you may not have to do much or get it built in from the machine.

 

Jimmy Carroll: [00:35:01] Yeah. Another question I wanted to ask. I know it’s taking these features and saving these files and not the actual images. But do you have any involvement or parallels or applications relevant to simulation and digital twins?

 

Roei Ganzarski: [00:35:22] Yeah. So that’s another interesting part. Actually, I was speaking on a panel not long ago and the digital twin came up. And I think digital twins are phenomenal. Done right, they allow you to try out things without actually having to do or harm the manufacturing part. And one of the things we were talking about, this was in the world of aerospace, that digital twins allow you to know exactly what’s on your engine. So all the parts that are on your engine, so you can know in real time, this is what’s on my physical engine and here’s the digital twin of it. And my response was, “No, it doesn’t.” The digital twin only shows you what the data told you is on your engine. But if I put a used part and gave you the wrong data, your digital twin shows one thing; reality is another. And again, that’s where the compliment is. We have to be really careful as an industry, manufacturing-wise, that we don’t get so enthralled by, “Ah, digital twin. That solves all my problems. Blockchain. That solves all my problems. Oh, AI!” No, at the end of the day, we all live, breathe, travel, manufacture physical items, right? So I may have this phenomenal data and a digital twin of my data, but if I did a physical swap and didn’t tell anyone, it doesn’t matter. Now I’m overconfident and bad things can happen, and you see it happening all the time, right? And so digital twins are phenomenal. We don’t replace the digital twin, but what we can complement is, let’s make sure that the physical part, let’s say on this jet engine to use that same example, is in fact what is represented in the digital twin. Because if they’re not, you could be making the wrong decisions.

 

Jimmy Carroll: [00:37:06] Yeah. That’s fascinating. It’s also refreshing to hear because there’s a lot of companies out there that do a specific thing, that tout – well, AI is a great example, especially in the industrial automation space. Seven or eight years ago, AI was being touted as a tool that’s going to solve everything. And now it’s a tool that’s adding tremendous value across disparate markets and applications. But, again, using industrial automation manufacturing as the primary example, it’s a collection of different components and tools that work together. And yeah, it’s like a complement. I can see how your technology specifically complements the existing ecosystem of industrial automation technologies. And to your point about digital twins, everything has its place, right?

 

Roei Ganzarski: [00:37:52] Yeah. And I think the key is for the manufacturers to first know: What am I trying to solve? And let’s find them the tools to do it as opposed to: Oh, I need to have AI in my factory. Why? Because it’s really cool and everyone’s talking about it. Well, that’s not really enough, right? You have to find out why. Are you having problems with traceability? Call us. Are you having problems with potential counterfeits or gray market? Call us. If you don’t have any issues like that or don’t need them, don’t call us. So from that perspective, it’s not about install it because it’s really cool. And it is by the way. I may be biased, but it is really cool tech, but it’s really about solving those specific problems.

 

Jimmy Carroll: [00:38:28] Yeah, yeah, for sure. That’s another point that’s been made by a number of people that I really appreciate. It’s like, figure out what you actually need instead of just assuming that you need this AI or this humanoid robot.

 

Roei Ganzarski: [00:38:42] Right, right.

 

Jimmy Carroll: [00:38:43] Why do we need humanoid robots? We’re not sure yet, but they are very cool.

 

Roei Ganzarski: [00:38:46] Right, right. Especially those dogs that walk around on their own. Those are super cool.

 

Jimmy Carroll: [00:38:51] Well, though. Yeah, I’ll save my opinion for another time. Those are very cool though. And they have real-world application. Anyway, Roei, anything else we haven’t talked about today that you’re particularly excited about?

 

Roei Ganzarski: [00:39:08] One, my team. What a phenomenal group of people. I love that,  and I will say that one of the things we run into after people stop not believing that we can actually do what we say, that the technology can actually, just with a picture, serialize things and identify them. So sometimes we’ll do like a Las Vegas magic trick. Hey, let’s get a deck of cards, open up brand new, and let’s identify them by their backs without actually having to look at what the number is. Then they go to the complete opposite. Oh, well, if I FeaturePrinted the back of the phone, I’m sure I could take a picture of the front of it and you’ll tell me what it is. It’s like, no, it has to be the back of the phone. If I fingerprint your right hand, you can’t give me your left hand and I’ll know what it is. It’s like, “Oh, but it’s AI!” It’s not magic. It’s not a trick, right? If you took a picture of the top left corner of your phone, you need a picture of the top left corner of the phone. Check deposits. If it tells you, take the picture of the front of the check and you take a picture of some other check, it won’t deposit. It’s that simple. And so sometimes, again, people have to understand: What are the expectations and what’s realistic? It’s really easy for companies to oversell what they can do and solve all problems and then figure out that they can’t. We are more of the transparent: Hey, this is what we do. If you take a picture of the top left, take a picture of the top left. It won’t work if you’re taking the picture of the other side of the product. If you want to be able to, take two pictures, of both sides of the product. That’s okay. But it’s really about setting expectations and understanding what you’re getting and why, to make it successful. Otherwise everyone gets disappointed.

 

Jimmy Carroll: [00:40:46] You know, I was going to close out, but now I have to ask the question. You’re saying that your technology could take a picture of a back of a card, and if you have 52 cards, they’ll each have their own unique FeaturePrint? How does that work?

 

Roei Ganzarski: [00:41:00] We have actually FeaturePrinted a ream of printer paper. And then we can identify which is which. Just a blank piece of paper, because again, so here’s the the premise of it. As you guys know, in manufacturing everything is made within tolerances. Machines cannot make the same product over and over again. That’s why we give it tolerances. I need that same phone. I need this phone to be 3 inches wide plus/minus two-hundredths of an inch, or 3 microns. It doesn’t matter. And as long as it’s within that six sigma of quality control within the band, we’re good, we’re happy, the quality is good. But it isn’t really identical. There are those small nuances of manufacturing, issues of manufacturing, flaws of manufacturing. It’s not a flaw in the product but of the manufacturing process that really leave their mark, a fingerprint. What we developed, FeaturePrint, is the ability to with a standard off-the-shelf camera, take a picture of an item and see those various aspects, flaws, and I’m talking thousands of them, and collect those into a digital fingerprint and say, That’s what makes that product unique. In fact, if I were to take two phones, both off the production line, right one after the other, the likelihood of having the same set of flaws or manufacturing aspects on them is about 1 to 3.5 trillion, which makes the feature print as unique as a human fingerprint.

 

Roei Ganzarski: [00:42:30] So it’s like a fingerprint in that it’s unique. It’s like a fingerprint in that it’s persistent. It’s not a sticker I can peel off or an etching I can scratch off. It’s the actual item. So it’s persistent. It’s always there. And, like a fingerprint, it’s inherent. I don’t have to do anything to have a fingerprint as a human being. I am born, I have one. I don’t have to do anything to have a FeaturePrint as a part I made. I have one because of the manufacturing process. And so from that perspective, we’re using that thing that everyone knows about, tolerances, except we’re able to do it with a standard off-the-shelf camera without training and without knowing what the part is. That’s why we have over 55 issued patents on our technology that makes it really well protected. There are a lot of companies who are trying to do what we do or claim to do what we do. Buyer beware, as they say. You may want to check when you’re about to sign with a company. Is it their IP or are they using someone else’s IP? One of the things we pride ourselves on is that everything was developed here in-house in Bellevue. We do not use anyone else’s algorithms, anyone else’s software. Everything is done in-house and protected from that perspective. And so you know that what you’re getting is the real deal.

 

Jimmy Carroll: [00:43:41] This is a really interesting concept. Like my son. Well, two of my sons — I have three sons — two of them go to school just down the street. Elementary school. There are seven sets of twins, and my middle son’s 5 years old, and he’s best buddies with this set of twins, and they’re virtually identical. So it’d be nice to have that app and just go up to them and be like, “Hey, which one are you?”

 

Roei Ganzarski: [00:44:04] That’s right, that’s right. And that’s the thing, right? Those twins. The one thing that’s different is their fingerprints. They’re very similar, but they’re different. And that’s how you can tell the difference. Same thing. A million parts that are identical actually aren’t.

 

Jimmy Carroll: [00:44:18] Well, Roei, thank you so much. This has been super fascinating. I appreciate the time. If people have further questions, I would direct you to Alitheon.com, or you can reach out to us at manufacturing-matters.com. Send your questions, comments, whatever. Happy to pass them along to Roei and his team. So Roei, once again, really appreciate it. Thank you.

 

Roei Ganzarski: [00:44:39] Thank you. It’s been a pleasure. It was a fun conversation.

 

Jimmy Carroll: [00:44:42] Awesome. Well thanks so much.

 

Dan McCarthy: [00:44:44] Thank you.

 

Jimmy Carroll: [00:44:44] All right. Thanks everybody.