The EDGECELSIOR Show: Stories and Strategies for Scaling Edge Compute

Artificial Intelligence vs. Natural Stupidity - Predictive Maintenance, Edge Computing, and Transforming the Automotive Industry with Hadi Nahari of Cognomotiv

September 19, 2023 Pete Bernard Season 1 Episode 6
The EDGECELSIOR Show: Stories and Strategies for Scaling Edge Compute
Artificial Intelligence vs. Natural Stupidity - Predictive Maintenance, Edge Computing, and Transforming the Automotive Industry with Hadi Nahari of Cognomotiv
Show Notes Transcript Chapter Markers

What if Artificial Intelligence could help predict and solve equipment issues before they even happen? Join us for an engrossing conversation with Hadi Nahari, the mastermind behind Cognomotiv, who is revolutionizing the tech world with their AI-driven solutions. Leaning on his expertise in cryptography, security, and reliability, Hadi unravels how his fascination for trustworthy systems spurred him to create a tool that measures the health of systems and predicts potential problems. We also dive into the evolving world of Edge Computing, its diverse interpretations, and how it's steadily transforming our lives.

The journey continues as we navigate the intricate world of the automotive industry. The conversation with Hadi unveils the challenges and opportunities that come with integrating his AI technology within vehicles. We discuss the rising number of automotive manufacturers, the need for consolidation, and the cultural, organizational, and technological changes needed for these companies to evolve into software-capable entities. The conversation takes an interesting turn as we explore the potential impact of AI models on the automotive repair industry over the next decade, including applicability to Pete's 1972 Volvo 1800ES.

In the final act, we explore the myriad applications of AI technology across various industries. We emphasize how data can be harnessed to detect patterns and anomalies, paving the way for preventive maintenance and diagnoses. Hadi sheds light on the importance of AI in bolstering technicians' capabilities, not replacing them. Join us as we navigate the future of AI in industries like automotive, and delve into the dangers of "natural stupidity". This riveting episode with Hadi Nahari promises a glimpse into the future of AI and its potential role in transforming industries.

Want to scale your edge compute business and learn more? Subscribe here and visit us at https://edgecelsior.com.

Pete Bernard:

When you ask people what Edge Compute is, you get a range of answers Cloud Compute in DevOps, with devices and sensors, the semiconductors outside the data center, including connectivity, AI and a security strategy. It's a stew of technologies that's powering our vehicles, our buildings, our factories and more. It's also filled with fascinating people that are passionate about their tech, their story and their world. I'm your host, Pete Bernard, and the Edge Celsius show makes sense of what Edge Compute is, who's doing it and how it can transform your business and you. So let's get started.

Hadi Nahari:

Focus do not disturb, so I don't get unnecessary blinks and not as important.

Pete Bernard:

Yes, the bloops and the bloops in the background usually are not important. All of that's fun.

Hadi Nahari:

That's what makes people feel special.

Pete Bernard:

Yes, we're interrupt driven in these days. Yeah, cool. No, I'm glad we could connect. Are you here in California? Is it a Menlo Park area or something?

Hadi Nahari:

I am in Mountain View, california. It seems like you're sitting in one of the Google's empty offices. When you live in Mountain View, it used to be massive, all the Googlers with the badges and flashes and ice builds and all that. Right now it's just empty Google offices. It kind of feels like when you go to Google campus. It's like a little apocalypse. So yeah, mountain View, california.

Pete Bernard:

Yeah well, it's similar. Up here in Redmond I would say we have a giant. I'm only about a block from the Microsoft campus and we have all.

Hadi Nahari:

I used to go every other month there, so I know you're there.

Pete Bernard:

Yeah, and beautiful new buildings Not really anybody in, not a good time to build a huge new campus, but it's just good for the local economy. It's like a lot of folks with hard hats on and orange vests running around, so that's good. Cool, yeah, I appreciate the time. Why don't we get into it? Why don't we let me introduce you, hadi Nahari? Thanks for joining. Sometimes I do these podcasts and I forget to introduce who I'm talking to and then like 30 minutes into it it's like oh yeah, maybe you can give us a little bit of a background on kind of your backstory and how you got from there to here or wherever here is, and maybe we can get started.

Hadi Nahari:

Yeah so, hadi Nahari, I am in some ways like it, you know, insecure, paranoid, you could say. I started my career as cryptographer, as a safety and security expert. It wasn't, you know, called cybersecurity I started that. Netscape.

Pete Bernard:

I saw that.

Hadi Nahari:

Netscape, all of the you know cool names that's called to our old to remember that name. It was the cool name back then. It was amazing environment and worked up and down the stack Sun Microsystem. At when AOL acquired us Netscape, there was this company called AOL, if you remember. Yeah, nvidia, morola, some of these big names. Really passionate interest has always been safety, reliability, security, all the above. When things are reliable, I feel safe. It's my insecurity is addressing. I love reliable people. I love reliable APIs. I love reliable investors, engineers, architects, marketing people, cars, hardware, softwares, static APIs everything that's my thing and kind of segue into you know where I am right now. I quit my very nice and high paying job After about a week, after I married my then girlfriend and started cognitive and I'm a CEO and founder of Cognom. What it really to just dedicate this passion to something that is tangible, which is making things reliable? We don't build hardware or software, autonomous stack, none of those things. We just measure the health of these devices. We call them equipment health and we use AI, we use machine learning, deep learning, like anyone does. It feels like saying it's e-commerce in the 90s Everything has to have E Right, everything has to have, you know, generative AI, that's right.

Pete Bernard:

Dot AI, dot AI. You get your domain name to be dot AI. That's what I heard it has to be, you got to have that somewhere.

Hadi Nahari:

So yeah, but really are really focused and what we do, what we're interested in, is to measure the health of systems, Equipments. They weigh from the human part. We just focus on the equipment and systems and we produce if this thing is healthy, if this thing is happy and if there's any problem. We detect and predict those problems and we provide recommendation as to what to do with it. Should be that this is okay, it's kind of okay, system is still, you know, operating or something you know, right, problematic. And then we provide guidance to technicians, right, what to do with it, fix it and all that. We don't fix the problems we find. Sure. In a nutshell, it's your McKinsey called it predictive and proactive maintenance. So we do that. And on top of that, one of our differentiators is we also provide guidance, so we don't leave the technician for technicians as okay, something bad has happened.

Pete Bernard:

Right, the engine light comes on and who knows it's like tough luck.

Hadi Nahari:

We're going to tell you something bad is about to happen. It's like brain surgeons they're really, really accurate and absolutely helpless. It's like they're going to tell you exactly in the 72 hours you're going to be paralyzed. There's nothing we can do. So we kind of don't do it that way. We accurately detect and predict problems, but we also recommend what can be done.

Speaker 3:

It's a thing.

Hadi Nahari:

That's where the AI thing comes in yeah. But really that's what we do, that's what our passion is.

Pete Bernard:

It's interesting. You mentioned that. You know it's interesting. I was talking to Stacy Shulman from Intel recently about the pervasiveness of edge computing. So we use this term, edge computing. It's kind of I call it, it's a bit of a Rorschach test. Everyone has their own sort of perspective on it. But she was saying, yeah, I mean you walk around and go into a hospital, look at all the equipment there, that's all edge computing. You go into the supermarket, all those checkout, self-checkout stands and the somehow, you know, maybe don't appreciate the ubiquity of all these devices and systems that we're now totally relying on on a daily basis. Like you said, they have to be reliable. Probably the people are the most unreliable part these days. And but how to get these systems reliable and predictable in terms of their behavior and maintenance seems to be, you know, kind of a tough, tough nut to crack, Right? I mean, there's a lot of things that can go wrong and I guess the one of the areas I saw from your website that you had been involved with a lot is automotive, and I think that's kind of a huge frontier these days. Talk about making sure things are reliable. Has that been consuming a lot of your cycles these days the automotive market, or where's the? What's the hot area for predictive maintenance and sort of anomaly detection and things like that?

Hadi Nahari:

Predictive, proactive, restrictive, all of that. Let me take a quick step back. What you mentioned is insightful, meaning everyone and their cats has got their own definition of edge computing. I remember Cisco came in when I was at NVIDIA and they said we got this fog computing. Oh, right, I remember that fog yeah, so you talk to mobile network operators, they, from their perspective, edge is the edge of the network.

Speaker 3:

Yeah.

Hadi Nahari:

They're a better people. And we didn't call it edge, we didn't call it IoT. We, when I was at Mono Vista, we called it just, you know, embedded computing. It's edge is endpoint. So, from our perspective, when I talk to non-mobile network operators, non AT&T's and Botophones on the world, we refer to edge as that endpoint device. In the case of the car or any vehicle, it's that little compute or many computes that exist in that vehicle. Right. In the case of factory floor robotics, it's like any compute that is in that you know anything that is not really cloud. Now you're in the cloud a little closer to the edge, a little deeper in AWS, google compute, whatever.

Speaker 3:

So that's for us edge.

Hadi Nahari:

I just tried to kind of consciously switch to now. These guys called edge the edge of their network when I'm talking mobile network operators, M&Os. So that's. I wanted to clarify that in this context. Yeah, we talked to the edge as that endpoint.

Speaker 3:

All right.

Hadi Nahari:

And I'm with you. We don't necessarily only work with automotive but as a you know 10 person startup, you're nothing but your focus. So we are very primarily focusing on automotive sector for variety of reasons. It's got a lot of you know addressable problems that the market has gone through a lot of changes. That from the business technology, all of those perspectives. It requires a lot of attention and the other part is up until like recently, every kind of fancy, fancy, advanced kind of work that was done on the edge was kind of scant work and science project in these R&D parts of these cars, except for this little company called Tesla that no one thought that they would make it. I'm talking about the company Tesla. I'm not referring to its you know CEO. I separate that. Mr Musk and everything. People who want to know what I think refer to my post on LinkedIn. I'm specifically talking about Tesla. The company Tesla, the automotive and everybody else was kind of looking at all of these advanced things and the car autonomy and everything Realistically, all the automotive vendors. It wasn't serious.

Speaker 3:

There was the R&D.

Hadi Nahari:

They had to have some sort of presence there. They got to do, show cool things and see us. But really, when you talk to under R&D with them, they're like all of this autonomy and all of that, whatever right Action happens in Germany and Japan and you know South Korea and everything, and they were not releasing cars with all of these sensors. It was nuts. Therefore, everything, why am I sharing this with you? In response to the reliability, let me bring it home, for everything that was the enemy of reliability was already very nicely addressed. The world was very reliable. We had this thing called.

Speaker 3:

AutoZar we have.

Hadi Nahari:

You don't touch things that are in the compute. You design something, you test the crap out of it for like two years.

Speaker 3:

Release the product, if there's anything that fails.

Hadi Nahari:

Bosch, denso, isen and others provided this nice DTC diagnostic trouble code and the systems tell you when they fail. You don't need to do anything else.

Pete Bernard:

It's all good. When the smoke starts coming out, then you know there's a problem.

Hadi Nahari:

And everyone would show that hey, take a look at our repair history. We are reliable, we don't need any of these things. To a certain extent they were right. Meaning, if you don't add any of these pendangles that we have been adding to the vehicle Jensen Wong famously called it, the vehicle is becoming data center on wheels, You're never, going to be data center on wheels. But Jensen said things like this when you got the third trillion dollar company on the face of planet, you get this thing like this and he's awesome. I learned how passionate a founder CEO is when I worked for him. In a certain way, it was right. Now, when you add all of this pendangles, software, sensor, lidar, all of the semi-autonomy advanced driver assist system, adas, all of those things Now that world starts to change really rapidly. The very reliability and safety and conservative sort of engineers of the world in Germany and Sweden and Gothenburg they started saying, okay, this is nuts, we're not going to do that in our vehicles. And they resisted for quite some time and one thing they got right. The other thing they got terribly wrong. They got it right that autonomy is not going to happen. I don't believe in my lifetime I'm going to see 40% of the transportation in vehicles fully autonomous. It's just not going to happen. I don't think even in 2030 it's going to happen, maybe in some restricted areas.

Pete Bernard:

On, tarmac Fixed route.

Hadi Nahari:

type of things Fixed routes, fixed area and everything. So they got that right. But the thing that no one got it right was electrification. And here's the cool thing, how that relates to sort of reliability. Oems realize that when they yank out all of that internal combustion engine, ice and transmission majority of the passenger vehicles don't have transmission all of that they don't put back just the battery pack and the electric motor. They have to put a lot of software. This thing is going to come in by definition with a lot of sensors and everything. They got to do it, otherwise they're continuing losing market share and with this unwanted push to advancing the tech, not towards autonomy, towards electrification, a lot of other things came in. Now they can't hire software engineers and better software engineers fast enough. Now they can't get their hands on camera, object camera and sensors and LiDARs and all those things fast and cheap enough. With that, the first enemy of reliability. Same thing for securities complexity. They are adding complexity. They are adding a lot of software and all of these things. Automotive vendors' software is not their forte, let alone data samples. That is kind of why. Long way to answer your question, tldr. Yes, it is consuming our attention a lot because there's a need. And also there's another thing it's a lot easier to control humans around the factory floor robotics, aircraft, engines it's once you add all of these in the largest IoT in the human history vehicles and a human is in it, around it, in the passenger and driver seat. When you mess things up in terms of safety, security, reliability, you're going to feel it. So that's kind of why, yeah, automotive's a hot.

Pete Bernard:

Well and also you have the fact that I mean, let's be honest, a lot of people aren't very good drivers either. So you've got not only kind of the safety issues around something moving at 60 miles an hour that needs to be safe and reliable, but you also have this person behind the wheel that maybe is not very reliable either.

Hadi Nahari:

So yes and no, yes and no. I want to touch on it, not be reactionary, but there is one thing as to you are correct. The assertion as to humans are not really good drivers In one aspect is really true. Take a look at death by drivers and caused by vehicles numbers. But there is one corner that some people have used, knowingly or unknownly. As to it implies, as if, machine is going to be better driver than a human. If, by implication, someone wants to make that conclusion as to, therefore, machines are really good, that's a really slippery story. Because the data suggests that machines are awful at doing this.

Pete Bernard:

Right, there's all those interesting paradoxes there. I think what you were saying about the automotive business I mean the sheer complexity of the systems. You've got a lot of interesting things going on there. I'm somewhat familiar with General Motors and some other things. My brother's actually worked there for like 37 years. I'm so sorry. I know the whole culture. There is very traditional culture and, like you said, you get something there, you get it to work and then don't touch it until it breaks. So I think there's been a really interesting cultural thing, very ripe for disruption. That whole market and obviously Tesla came in and did a lot of that. But now the systems, like you said, these are not software companies but they've had to very quickly become software capable companies. Think about software development, deployment, release, maintenance, all the things that a lot of other companies have been doing for a long time. So cultural changes, organizational changes, technological changes, and it's a huge market. Um, I was talking to someone who is in China recently. I don't know if there's over a hundred automotive manufacturers in China alone 450.

Hadi Nahari:

Yeah, I mean it's incredible yeah.

Pete Bernard:

Yeah, I mean obviously, yeah, they'll have to consolidate there. But uh, yeah, it seems like it's funny. I'll tell you a little anecdote of a very non technical failure. So I have a, I have a 72 Volvo um 1800. So it's like this little tiny little coop and um amazingly reliable. That those very reliable. Although I was feeling fourth gear one day, it was like a little crunchy. It was a little crunchy.

Hadi Nahari:

Come on, you gotta give it to me.

Pete Bernard:

And it was getting a little crunchier and crunchier and ultimately I had to replace the transmission, so that so my anomaly detection was like my right hand feeling that could the crunch of the gear we're getting a little, yeah, and it was, it was, it was definitely a reliable system to detect anomalies. Um, unfortunately it was a multi month uh uh effort to kind of get that thing swapped out. But uh, yeah, no, I like driving the old cars too, because you have the uh pretty analog experience there yeah. And you are viscerally connected to the road. But the cars today, yeah, and automobiles whether they're, you know, like you said, commercial vehicles on fixed routes or these, you know, personal vehicles uh, the amount of systems there that need to be upgraded and really kind of thrown out and replaced is tremendous. So when you guys look at, uh, a system like that, I mean as your stuff deployed, like throughout, like the CVC and some of these other components and the and the IVI, et cetera, where you, you know there's like the I don't know if people know there's these different subsystems within an automobile, like the central vehicle controller and uh, all that stuff, are you focusing on? Um? And also, I was curious, like, how does that connect back into the cloud? And are you you're running like AI models against some of this data that you're measuring and kind of in real time, I assume? Right, this has to be pretty deterministic, great great question.

Hadi Nahari:

Great question Um. I will just put all of this in in context very quickly. Let me uh very quickly make a you know uh plug in and uh comment on, uh, the GM's of the world, gm for Mercedes and others. There is one thing and, and it's very easy to kind of uh, uh, I don't know, criticize them by hey, they're like you know, not advanced and everything, and some of them they deserve any income and deserves it. But keep in mind uh, especially when it comes to uh, you know, such a sophisticated and advanced and complex system, which is a system of systems like a vehicle. It's the most complex um device that humanity has created in at this scale, way more complex. There are 16,000 different components on average, from the plastic you know uh, log, nuts and everything in a vehicle, all the way to you know very uh uh. You know complex uh, lidar, things like that in a vehicle. There is one thing to build a single vehicle that kind of um goes in a Stanford around the track and does something really fancy and create a beautiful uh video and then raise a hundred million dollar, captain and Burnett, in two years. There is another thing to be able to pop out at every 60 seconds right, pop a vehicle out of manufacturing, right At scale, and satisfy the um believe it? The complex supply chain that expands from the parts of China all the way to, you know, western and Eastern Europe and all that and maintain the same. You know qualities, you know ISO 9,000, you know and all that you know year after year, model after model. You got to give it to the GMs and forwards and others and one of the magics that Mr Mosk did with you know, in such a short time, with, uh, Tesla still catching up qualities crappy in their product, uh, but that's a different issue. Uh, it's, it's a different thing. It's similar to AI and I come back to your point quickly. Similar to AI. Assembling something that is really impressive on a website, really really easy. But creating a product out of it which is, you know, repeatable, which really addresses the real problem on mass, on scale, is a different uh, uh, you know phenomenon. So people are investors, entrepreneurs who are in this field. They understand the difference, massive difference really, really, uh, uh well. So I wanted to just mention that, uh, you know it's, it's easy to criticize those large companies, but once you uh go to a factory floor, once you see, uh, what it takes to assemble this and how complex, just you know, a a floor from end to end is, it gives you a different, uh you know, appreciation as to why they resist some of the changes, because change is bad. To answer your question now, what do we do? The vehicle, as I mentioned, is system of systems. It's not just one computer, one chassis, it's. There are many, uh you know, electronic, uh uh controller units, ecus, little tiny computers with different capabilities in the vehicle. An average, you know, between 62 advanced vehicles, maybe 160, 70, uh of them. They're different connection, you know uh networks. They call them can box or flex tray, or these are like little networks that these connections are connected together. Usually on average you know, between eight to 10, 15, some of them rocks, have more, uh, each one of these computers. Some of them are, you know, micro controls, very you know uh, small things, they just move the window up and down. Some of them just, you know, move your uh, windshield wipers. Some of them are more sophisticated, they are full-fledged computer, like a uh in vehicle infotainment, which kind of is a misnomer these days. It's not just for infotainment, it's actually a central, you know, computer, as you mentioned, cbc, things like this. Uh, there are some, uh, you know some of these components that are less sophisticated, some that are more sophisticated, Right, so I'll answer your question. We do a couple of things. There are a number of configurations. One of the things we do is when we are embedded in the vehicle, sitting in one of these systems, we really don't care which one, so long as there's access to data, um, where data monkeys usually that component is called connected gateway where it's kind of like a gateway, sits in between many of these connecting networks, right, and kind of sees different traffic. Sometimes we sit there, our customers embed us there. Sometimes there is, uh, you know, we kind of they already have their device called telematics, a lot of you know fleet. They have this little device, that telematics kind of, installed in the vehicle, right, you know measures, a lot of things. Sometimes they put us there. So we have this edge AI component I come back to it which sits there. That's one configuration. Wherever in this system of system we sit, um, that configuration is embedded. Um, pros and cons, once you're embedded, things that you mentioned, like, uh, you know real-time-ness, anything that happens right there, and then you see it, uh, you won't encounter the case that I go to the mechanic. I might this vehicle really spotted like 10 minutes ago, so I don't see it. So, you see all of those things. Yeah, uh, you get to kind of assess on the fly things like that Uh, but you gotta some someone has to put your uh software there, so it takes a you know uh, getting to the customer, getting to the end point, has got more you know business challenges.

Pete Bernard:

A lot of integration too.

Hadi Nahari:

You gotta be integrated. So, but once you're there, we have the number of these uh for, you know, our customers, uh, things are really good and you've got like real time. It's like going around with your own X-ray and MRI on your back the moment something you know uh happens. You are able to kind of see did I just run fast, I'm, I'm good, the heart rate is, you know, going up, but I'm good, right, right, or am I about to have a heart attack? So I got to go to uh, you know, uh, to uh ICU. Things like that Become apparent in the cases that you are having safety critical or impact to the equipment. Is equipment is very expensive or impact on logistics, whatever. It makes sense to uh have uh, this uh, you know, embedded in reality. In about 10 years from now, all the vehicles will have a version of this embedded uh you know, cognitive or or not, in the system. It's just the right thing to the right thoughts, a matter of logistics and getting it into the integration flow. That's, you know, three, four years for difficult uh manufacturer. That is one configuration.

Speaker 3:

Yeah.

Hadi Nahari:

This configuration is not our de facto configuration because of the business challenges, integration challenges that we mentioned. The second configuration is remember where data monkey, many of my customers, fleet OEMs, uh, also dealer networks they are ready have collected data from the vehicle. They have access data from the you know vehicle. They have different type of data from and about the vehicle, textual data. Someone says this vehicle has got a little you know vibration in the front, you know part of the chassis, or I see red engine light coming up. This is just textual. That's called you know, a customer description or a technician takes a look at it. It's like I think we got to take a look at this recall item. So this is textual about the vehicle. They connect the probe and they capture freeze frames. Something bad happened and ECUs usually you know store that kind of screenshot of the traffic and capture that is freeze frame and number of other. You know different. You know flows from the view, from the vehicle. So they already have it either at the point of service or you know about the vehicle what they do. In this second configuration we're not embedded, we are in the cloud. They just plug in all of that, throw it at our system.

Speaker 3:

You are magic, right.

Hadi Nahari:

And we tell them, based on this data and this is, from you know, just the textual data, all the way to all of the sensor and raw signal data that's part of our differentiation we're able to tell them what the problem is, what are the parts and labor required for this labor codes and parts code and everything and what they need to do, what is the you know procedure to fix them things like this, right? So in this case, we are using the same back end, the same exact cloud part, but not having direct real time access to the vehicle. From our perspective and I shot up after this from our perspective, really, it's the same thing. Our models are the same. Our flow is the same. Sure, how frequently you get the data is really the difference. If I'm in the vehicle, our model is I'm getting a flow of this data continuously frequently and the data recency is high when I'm not there. Every month for the fleet, every week for the you know rental cars, every you know two, three days on average, you know I get this data. The task is the same. It's just exactly how. What is the data from one data point to another data point? Yeah, that's what we do.

Pete Bernard:

The richness of the data. It's sort of like so so that's pretty cool. So you've got sort of the embedded version, which I agree is kind of a. It's a bit of a lift to get that integrated and deployed and it takes many years. And then there's this other one where it's sort of like it sounds like it's sort of like an AI powered Mr Goodwrench or something like that, where you know a mechanic or someone can kind of.

Hadi Nahari:

We call it Kyra. Let me plug in Kyra.

Pete Bernard:

Kyra.

Hadi Nahari:

AI repair assistant.

Pete Bernard:

Kyra Okay, good, that's a good one. I can remember that. Yeah, no, that's cool. So it's sort of like now you're taking AI and the cloud powered AI and looking at like time series insights as well as, like you know, like you said frame pictures of the data over time or in an instance, and then you're probably hopefully I don't know, it'd be interesting Are you comparing it to like other car, like the same car that you know, the same model? I mean, is there that kind of like intelligence yet out there where you can say, oh, this is a 72 cor, there's a 72 Volvo, it's got this issue. All 72 Volvos, if they have this data, that means you know the transmission is about to die.

Hadi Nahari:

Just to go back to my example, so up until right now really what we have been discussing was, you know, data acquisition. Yeah, and as data acquisition systems have existed, you know, for many decades, it's like we just talked about you know you're talking about the ODB, like the ODB port on cars, like OBD2 onboard diagnostic. Yes, that provides data acquisition. Direct connection of the probe is data acquisition. When some technician or customer calls in it says my vehicle is not really operating well, that is data acquisition. All of these are different modalities of the data. You take a picture of the you know smoke and read out a you know sensor data, and all that through OBD2 or not whatever. But all of that is in the. Really, you're just data acquired. Now, the magic now is you know AI happening from here. Up until now, you had business analytics. You had you had business intelligence. Remember the days that calling artificial intelligence AI was. It was a taboo right. The first three generations of AI. A lot of people promised a lot of things. It was a bad thing, yeah, and we're kind of about you know a half way there in this generation as well, yeah, but I'm digressing. So up until now, we talked about you know data acquisition. The magic right now happens in what you just said. Now, with all of this data, with the recent you know current systems that we have advancements and understanding and processing textual data. Understanding and processing Another textual data is also similar to time series. This is a form of time series. When we talk, it's in a sequence of the words and letters that we use. Another you know raw data, time series, you know Canvas and everything stored. You know data like the freeze frame, like I don't know rare event numeric data. You know pictorial data, all of those things combining all of them, all of them together, which is, you know, multi modality that everyone is right now just looking after. So we have a solution for that. And another thing, which is, you know, we have this other solution is called Chai cognitive hybrid.

Pete Bernard:

Yes, I saw that that's another acronym. You're good with the acronyms.

Hadi Nahari:

Yeah, it's like when you go to a doctor. You kind of like acronyms. When you go to doctor, the doctor doesn't go to a medical school just for you. Right has gone to medical school, has, you know, seen many patients similar to you and is able to extrapolate, is able to kind of learn from one patient throughout the residency and apply part of that knowledge to another patient. So systems that we have do the same thing, meaning I don't need to necessarily create, you know, model per vehicle. That's just death by you know 1.1 billion cuts. We get to do some cool things, which is, you know, transfer learning. You know I learned from one system I applied to another one. Another cool thing, which is in order to, you know, learn things, because our systems, any AI system, is a learning system. Instead of being told what to do, it learns what to do. That's really what is happening at a high level, and the way to learn is a lot of this data is. More than 95% of the data is thrown away at the edge. A lot less than you know. 5% of the data that is generated every second finds its way to the cloud. So a lot of this data which is in these edges, in these equipments are kind of generated time series and all those signal and everything. They're just dropped on the floor on the road. So what we do is having these edges, having these systems. We're able to not only learn from one applied to another, not only learn from a fleet and you know cluster of systems and apply to one specific system. Not only you know, increase the accuracy and everything. We just need to see events only once. So in the beginning we're not as smart as a shop foreman, but our systems provably and you know, measurably very quickly learn things in a very you know machine. I'm just raving and bragging about the AI. Sorry, I'm next to a buffet.

Pete Bernard:

I see, I hear an airplane there.

Hadi Nahari:

So, and that is when the actual, you know, machine learning.

Pete Bernard:

Yeah, that's the magic there.

Hadi Nahari:

Now that is where we come to picture as to the real value is there, and I tell people as a joke, if it's on PowerPoint, it's AI, if it's on Python, it's machine learning. So it's really very important. Yes, yes, no.

Pete Bernard:

I think that's amazing and when you start to sort of, like you said, the collective learning you know once you start seeing, you know, multiple vehicles of the same model or same symptoms, I love the idea of some sort of aftermarket system that people could, you know, use to sort of keep track and, you know, do predictive maintenance and diagnoses of their vehicles. Not sure if it would be super applicable to my 72 Volvo, but because I don't have an ODB port, but you know things like that it's fascinating, while 72 probably I don't think it was an ODB on that one there, but no, I is. What you're saying is like, basically, the data is coming in lots of different forms, right, like you said, it's sort of visual, it's digital, it's all kinds of things, and then, you know, the magic is really, how do you, how do you kind of synthesize and look at that data to find patterns or anomalies to patterns that say, hey, there's a problem here or there's, you know too much current on this or whatever that's going to be. You know something that's going to fail in the future or maybe something needs to be looked at, and that seems to be sort of a universal thing, right, it's like, I mean, any of this equipment would really need to be, you know, monitored, either in real time or post facto, to kind of gather that data. And, of course, every time you're analyzing that data, you're making the system smarter because it's learning. You know these patterns over and over again. So, yeah, no, that sounds, that sounds pretty exciting, I think you realize one thing, if I may, everything that I mentioned.

Hadi Nahari:

Going back to the first question you asked, everything that I mentioned is as is applicable to other embedded systems. There's nothing specific about the vehicle except for scale, and that is part of the reason that we don't say cognitive is just for automotive. It's this methodology, this way, this style of applying technology to create assistance for technicians, not replacing them. We don't replace our my first class citizens are the blue collar technicians. That's, that's that's I know who we cater to. This applies to all the technicians, because they're overwhelmed by the advancement in technology. They are really quality driven and they're financially motivated to do a quality job. If the comeback which is you get your car to fix and you pick up the car in day or two and either, that problem is not solved or it caused another problem. That's a terrible thing for them. They don't get paid by that. It's a bad reputation and all those bad things happen. So we eliminate the comebacks. We help them do their job and we make them. You know, help them make more money. We tell them our motto is do more, make more, and that is not specific to automotive. That is very pertinent to any.

Speaker 3:

Sure HVAC repair, I mean any of that stuff maintenance repair.

Hadi Nahari:

I could say the same thing about. You know MRO maintenance, repair, overhaul. You know operation of aircraft. It's much more regulated, but the concept is the same. So our job is to use data right to create an assistant, which is Kyra, and provide them guidance, detection and prediction problems you know diagnosis, prognosis and also help them what to do to fix it.

Pete Bernard:

Yeah, and it's sort of like, and I think that's kind of where you know people, you can go into a whole AI thing here, but the people get freaked out about AI replacing. But really AI, when it's well done, is super charging people's capabilities, you know, and giving them super power and recent moments.

Hadi Nahari:

Mark Andreessen last week just figured it out what we have been doing for years with Kyra, which is, oh my God, the best you know use case. I'm saying it jokingly I got great respect for them as a startup founder. Never poke fun at VCs. Yeah it's, it's like this. Assistance is one of the most prominent use cases of AI. Whoever said that the AI is going to come in and, you know, replace humans was either you know, ignorant and uninformed or they had other intention. This is just a stupid thing, we do it day in, day out. I can tell you we are generations, not just one generation, generations from any AI being able to replace any part of the human. I'm not worried about that. I posted this on the on LinkedIn a couple of days ago. I'm not worried about the dangers of artificial intelligence. I'm a lot more worried about the you know dangers of natural stupidity.

Pete Bernard:

Yes, there's the natural stupidity paradox. Yes, I agree, but yeah, no, that's interesting. I think that that's definitely like Microsoft uses the term co-pilot, which is a nice I think that's a nice phrase and then in GitHub and then on now in office and other things. But it's like, yeah, how do you, how does AI supercharge or give you some superpowers to do what you're doing better and leverage? You know your skills, especially like mechanics and stuff. I have a guy go to and Ballard, who's all he does is vintage of Olvos and he's kind of the Volvo whisperer and you know the ability for him to with his hands you can't see because I'm not on camera here, but you know the hands on skills are incredible, and so to be able to supercharge that with with more knowledge in the background is pretty key. So, yeah, totally cool. So you know, I don't want to keep you too long for your flight, but any, any kind of closing thoughts on what we talked about and what people should be thinking about when they're thinking about reliability and anomaly detection, as I like to call it.

Hadi Nahari:

Yeah, there's. You know we could go on the technical. You know parts of anomalies and not every anomaly is a bad anomaly. I tell people that you know you go down, you know one on one or I, 95, whatever, the first time that your vehicle is going 100, 10 mile an hour, that's an anomaly, that's really bad.

Pete Bernard:

That's a heck of an anomaly, yeah.

Hadi Nahari:

It's an anomaly, it's just. It's the first time that your system is done it. It's anomalous but it's not a bad thing. So right now, my detection is it, you know, prerequisite to you know batting not all the anomalies are bad to be able to perform anomaly detection and also then identify is it a bad anomaly and good anomaly? That is the real thing. Yeah, I just being happy that I identified anomaly. It's not really useful. The other thing, which is, you know, kind of, on the one side, it's a bad news because it's like, oh, we haven't solved that problem. On the other side, it's a good news as to how limited these systems are. I tell people that you know there is one side of data science and AI and machine learning and all of this jazz, which is, you know, we get this data. We don't tell people in our you know medium articles a lot of blogs are like this which is this data is really nicely tabulated, very clean, perfectly labeled, beautifully organized, and then this, this writer you know, runs a metric and algorithmic code and first time it's like shit, this is not working. Really nice, you know, modifies the data. You know some apparently, some presidents in Stanford did the same thing on their data as well. You do a lot of you know massaging and polishing and modifying and everything, and it becomes beautiful and it shows a look. I'm able to take place based on the rate of your typing, who you are better than your fingerprint, and people you know start losing their you know what, as though my AI is just taken over the world. What they don't see, what is not written, is how much work is required, what are the prerequisites? How clean this data, how theoretical all of this is? and how unrepeatable many of these things are, and I come across many of these claims day in, day out. On the one hand, it's like you gotta, you gotta really not be worried about the eyes taken over were like generations from that. Whatever works has been really so far helping, whether it's in the edge or cloud, I don't care has been really on the helpful side force for some good reasons. Because in order to apply any technology machine learning, deep learning, analytics, ai, software, whatever you want to call it you have to, you know, address a real problem, not just come up with a solution, for problem doesn't exist, it doesn't scale. You have to have something that is repeatable and you know welcome to capitalism it has to make money. Many of these things are theoretical and are designed to steer fear or designed to get clickbait and everything. So, look at it, you know, with with a little bit of you know big rock of salt and see if any of them has in reality gotten anywhere close to the real world. The data in the real world is nasty, it is unbelievably noisy. It doesn't come with labels. The data that is running in your canvas or the Tesla's canvas doesn't matter how old or new the cars are doesn't have a nice label as a bad data. I'm a good data, so you can't run you know supervised machine learning, things like that. The data is unbelievably high rate. It's like a one autonomous vehicle on average generates one terabyte of you know signal, raw, nasty data on an hour of driving on average. We don't even have processing ability to process that much data on a you know big ass server, let alone on the edge device. So many of these problems to the art that people are scared when I talk to them are really theoretical. But they do see, like you know video on TikTok or something and they get scared. And I'm not that scared. I'm still much more scared than natural stability, than artificial intelligence. That's my closing point.

Pete Bernard:

I think that's a good way to close it. I think you're right it's. You know also, like you said, working back from the actual problems being solved and you know what's the real value for the business or the customer. And then, like you said, the reality on the ground, with all this noisy data and stuff, is it's complicated and no one should underestimate that and and try to avoid the TikTok stuff. But yeah, I hear you.

Hadi Nahari:

Cool. Well, no, I look at it and have fun and enjoy and move on.

Pete Bernard:

Exactly, yes, it's entertainment, good entertainment value. Well, heidi, thanks a lot for the time. I think it's been super educational for me, hopefully for the listeners too, and it sounds like you're really in a fascinating space and a lot of probably interesting projects ahead too, so hope we can keep in touch.

Hadi Nahari:

I look forward to it. It was fun chatting with you. I really appreciate your time and attention and very open to chat with you again and hearing about you know any of these comments from your listeners and I look forward to your podcast and it's very informative and I look forward to continuing the conversation with you and listeners going forward. It was really fun. I appreciate it.

Pete Bernard:

Sounds good, all right, take care, take care you too.

Speaker 3:

Thank you for watching, thank you.

Edge Compute & Predictive Maintenance in Automotive
Automotive Industry Complexity and Disruption
Vehicle Data Analysis and AI Solutions
AI Assisting Technicians in Various Industries