Vita Koreneva (AlphaMille):
Welcome to AlphaMille’s Top of Mind podcast with Daniel Morozoff, and let me please quickly introduce him to you.
Daniel Morozov is a co-founder of VirdroVR. Some people say VidroVER, but it's correctly VidroVR.
VidroVR is an innovative machine learning company with a mission to help companies create actionable insight and awareness through multimedia understanding. And today, it's tackling some of the world's most important problems, like disinformation or information gathering, I would say. So, it’s analyzing a big scale of videos and investment opportunities, and really making sense of unstructured data bid for banks, private equity firms, hedge funds, or any other entity that cares about working efficiently and effectively.
Daniel went to UCLA for undergraduates to study physics, where he received his degree in neuroscience. And then he moved to New York and pursued his PhD at Columbia University; he had a project in the university that he worked on, which really became the foundation for VidroVR.

Daniel, welcome. Thank you so much for joining AlphaMIlle’s podcast today. We're excited to speak with you. And who can better present himself than you? Would you please talk to me and tell everybody about your own story and your beginning in this exciting space of AI?

Daniel Morozoff (VidroVR):
So first off, it's great to be here. Vita, thank you for for hosting. And thank you for kind of reaching out and getting us involved in this. I'm excited to be here. And again, big thanks.

To answer your question, How did you know how did I personally get involved in building kind of machine learning and computer vision or probably AI tools is we had a collaboration projects when I was doing my PhD at Columbia I worked on, as you said, modeling that kind of overlapped, neuroscience and computer science. And so I worked in both departments, AI and in the Computer Science department; our project basically gave way to the development of the foundational technologies that would, you know, turn into a company as it is today. And, you know, the problems that we were dealing with, then are actually very similar to the problems that a lot of people are dealing with now and kind of give credence and give validity to our business.

Effectively, those are problems having to do with unstructured data and trying to figure out ways to create relationships or create structure around that unstructured data to drive business or operational value from from that data. And that is what VidroVR is,- it's a set of tools or toolkit of machine learning and artificial intelligence algorithms that take unstructured data, you know, like audio, video and image, and then, basically enable users such as organizations and businesses, or primary customers or businesses to utilize that those unstructured data assets to boost the performances of their businesses. And because we're a horizontal technology, we basically serve a number of different user types. So we have people who work in the media sector, we have people who work in government, we have people who work in financial services, and all of them are using our tool sets to serve their specific business needs, but all of them share the same problem with unstructured data.

Vita Koreneva (AlphaMille):
So, VidroVR essentially is able to track information across the web like all kinds of video, audio outputs and text?

Daniel Morozoff (VidroVR):
Yeah, so, we primarily focus on, the unstructured part of data so we built an engine and a platform that basically takes any type of input feed, be that a video on like broadcast or social media, audio on a radio station or hosted on a podcast like we are having here right now, or imagery that's basically sitting in a database.
And we take that information, and automatically process it. We use whatever machine learning models are required to process that data type. And then, we extract the meaningful information from that for the business. We structure that data on, and then, from there, we basically create the next step, which are kind of secondary systems. So, we have are a no code workflow engine called actions, that allows you to take all that information that we structured and automate against it.

So, let's say you do some task every day with an unstructured data source, right? Let me give you an example. I want to understand market movements based on how a certain consumer goods is being talked about or sold, or discussed on social media. How can I actually as a financial services company, how can I tie those things together? How do I process that information? So our system is able to go out, collect that information, process it and then, tie it back and provide you the signal? What products or what problems are those products having, and then tie that back to, you know, in this case, an equity stock or or whatever it is.

Vita Koreneva (AlphaMille):
It appears to have a really wide application; you mentioned media companies, being able to gather information, also have a very effective overview of video outputs and not spend as much time doing so. They are even able to cut the analytics time or the viewing time, right?

Daniel Morozoff (VidroVR):
So, it's not so much viewing time, right? And media companies, and actually, across the board, like the value-add that we bring, is we're basically a technology enabler, right. We enable human beings to be able to perform better and to be more efficient at the tasks that they do. And we are able to do that using artificial intelligence technology.

So effectively, AI in our in our case, is basically the means to an end, right; we're an AI company, because we don't know any other way to do this any better. And so we build AI models to solve these problems. And in the context of media companies, we effectively are able to significantly improve their day to day operations in terms of increasing four to 500%, in performance against normalization of a human being.
So let me give you an example. Let's say, I'm able to do something in five hours, using our AI tools; we can get you down to one hour. And so when we do these comparisons, this is how we sell our company, our customers, we basically present them like ‘hey, this is what you're doing today, now do it with our system’. And they see this huge shrink in the amount of time or effort that's required. And so they incorporated AI technology, and become more efficient in the process;, and in media companies that relates back to say publishing content, or creating residual content, or better organizing various types of multimedia content within their organizations, which is their core product that they sell, right, that content is what media companies produce and sell to advertisers and consumers.

Vita Koreneva (AlphaMille):
Oh, this is massive; it seems to be providing an incredible competitive advantage to the companies that use your tool.
Why don't we talk about financial markets and financial companies? Because that's our core interest.
Could you please mention what use cases you had that have already found success in the financial arena?

Daniel Morozoff (VidroVR):
Right. So, I mean, in financial markets look to any part of an organization? I think it depends how you look at financial markets is a huge kind of industry and sector, right? So there's all sorts of things where unstructured data is utilized across the board, right? So you can look at scenarios where you have everything ranging from retail banking scenarios, all the way to investment firms, hedge funds, private equity shops, and each each one of those has very different applications for unstructured data. For example, in private equity, right? The idea is like how do I basically better compete in with my portfolio or my port coasts, against other companies in the space, right? So in that context, it's very similar to any business that operates in that industry. And they can utilize our tools to understand kind of the competitive landscape to understand how their products are better served or engaged with, or how their competitors products are being better served are engaged with and effectively adapt accordingly, effectively lowering their costs, improving their efficiency as a business, and then effectively performing their fiduciary duty as a private equity firm to sell the businesses that they invest in at a better valuation and make money that way. And that's very similar to hedge funds that are basically doing trading strategies against whatever they're trading right. In terms of like retail banks, again, if you have customer focus sectors, we've seen everything From pay, I have a lot of unstructured audio data of like customer service logs that people just calling in asking all the way to, to folks trying to prevent skimming operations on ATMs, to creating multi factor authentication for bank account logins, right. So like, I want to create a way for my customers to feel more secure, and provide another layer of security by creating kind of a new multi factor authentication regime using unstructured data assets, such as audio or video data, right. And so effectively, that allows you to say like, I want to opt in, every time I go to an ATM, not only does it allow me to type in my pin, I also want it to recognize my voice or to recognize my face, or my family's face or whatever, right, like you can create those rules. But in order to do that, in a banking scenario, the bank needs to have ml or AI systems in place to deliver that capability,

Vita Koreneva (AlphaMille):
what would be a very narrow application for those companies that are interested to track their investments?

Daniel Morozoff (VidroVR):
Again, it really depends what you're looking at, right? So like, a good example is, let's say you're trying to do a hyper localized investment in like a region, let's say you're interested in mineral resources or something like that in the middle of South America, or in Africa, or in the South Pacific, whichever, right? Fundamentally, the problem that you have, as an investor or someone who might become an investor is information acquisition, and then information processing. And in today's world, that problem is exponentially more difficult than it used to be right. And so historically, the way people have solved this or approach this is you either send people out into the field to collect information for you on your behalf. Or you buy information from data providers or research firms, right? Folks like Bridgewater, and like you can have a large number of them, right? The fundamental problem is that those firms are still doing the same basic task, they're going out there trying to collect data, and they're trying to buy it, and then they're trying to effectively structure it for you. So it's digestible for you to make some kind of investment decision. What we enable you to do that we enable you to shortcut that entire process by basically saying, hey, I want to be able to go in anywhere in the world where there's data available to ingest, I want to take that data, I want to structure that data without actually knowing what the structure is beforehand. And then I want to be able to extract signal that's relevant to my investment decision, and then use that signal to inform whether or not I want to invest or not, right, so an example being let's say, I'm trying to invest in mineral somewhere, I can tap into, you know, radio stations, brought local radio stations, broadcast stations, any social media in the area, create a holistic map of what people are saying, what a visa be mineral resources in kind of that region, and understand what is the outlook based on kind of the hyper localized information that I'm receiving from the ground, around how mineral resources are being applied. Now, if you think about that, you're like, wait, but if I go somewhere in the world, you're speaking a different language. Yes. And our system automatically handles that for you. So you plug us in, we take that data, and we we transcribe it, we translate it, we then extract meaningful information like instances, visual instances have saved mining operations being discussed, we then identify those segments of either video, audio or image, unify that in a holistic kind of structure, which is in our case graph. And then we allow you to receive a report against it. So you say, every time someone talks about minerals, are they positive or negative? What is it exactly that they're talking about when they're talking about minerals? Oh, interesting. There's a Chinese company that's investing in real, you know, mineral extraction in this region. Oh, you know, there's a US company that's involved. And so those types of things become very important for you to make your business decision. Because depending on what type of investor you are, where you're coming from, and what your portfolio looks like, already, all these details matter. And the more of these details you can receive that are basically focused in on your specific decision making process, the better your, you know, your process becomes. And so we can shortcut that for you and kind of give you that information on that data and effectively make your investment decisions better.

Vita Koreneva (AlphaMille):
Understood. So So that's great. What if I wanted to track a specific person mentioning lithium or gas, you know, any other person of interest from my business?

Daniel Morozoff (VidroVR):
The way we operate is on this notion of feet. So you put us in you plug our system into whatever data feed you want. That could be Elon Musk's Twitter account right now, or it could be a lithium refinery or lithium or company somewhere in in South America, it could be a lithium battery producer somewhere in China, it could be Tesla in the United States, right. And then basically, what you can do is you can create a way to monitor the entire supply chain and receive signal from every part of the supply chain. And so if you're interested in trading Tesla, and you're following kind of their earnings calls and quarterly reports, you'll know that they're trying to verticalized that supply chain. And to be able to understand the impact of say, you know, the global pandemic on that supply chain, before it actually gets reported is a huge win for you as an investor, investing Tesla, in Tesla or driving Tesla stock,

Vita Koreneva (AlphaMille):
we're sure it's fine. Time is of the essence, glass, how many languages can you do it in.

Daniel Morozoff (VidroVR):
So today, we support about 100 languages, on transcription, and about 80 languages on OCR, which is basically optical character recognition. So we read the text on screen. But a lot of those 80 character 80 languages are actually overlapping languages. In other words, your alphabets are the same. So like English and Spanish have a lot of similar letters except for a few. So we don't need to create a new specific language. And so basically, we have a smaller set. So that's, that's basically what we're doing right.

Vita Koreneva (AlphaMille):
Gotcha. Sometimes I understand you can even be faster than media companies, you don't have to wait for the reports, because whoever is using your tool will be able to get essentially information firsthand,

Daniel Morozoff (VidroVR):
right? If they plug into the right feet, right, so if they plug if they're monitoring the personal, let's say, like, case in point, you're monitoring someone on the ground, and they're recording something, and they're pushing it to social media, or they're streaming directly to yours, to your service running our platform, then yes,

Vita Koreneva (AlphaMille):
yeah, yeah. So it has to be a public or it has to be a channel that you can tap into essentially,

Daniel Morozoff (VidroVR):
or what Yeah, exactly. It has to be accessible. Yeah.

Vita Koreneva (AlphaMille):
Well, now, in terms of your clients, what is the size of your clientele? Because it I guess it directly leads into the question, How expensive is the tool?

Daniel Morozoff (VidroVR):
I mean, so our clients range and scale, right, we we actually have footprints of like small businesses all the way to kind of federal government customers, right. And so we we've seen, we have various deployment types and variants that serve different sectors and or sectors prefer, because of the data scales that they're dealing with. So because we deal with different sized companies, we also have different costs associated with different deployments. So for example, we have a SASS platform that you can get started for like basically starting off very cheaply, all the way to kind of an on prem deployment where we're basically running our software on hardware that either we buy, or you buy, and we deploy on that hardware, and then you manage the hardware. And we basically serve as our software running on that. So like that entire distribution we support. And the prices vary based on those, but in our primary customers are b2b enterprise. And we, we work basically in the sweet spot of b2b enterprise contracts on about like, I think right now that about 200,000 180 to 250,000 per year, contract value. And that's pretty much what we've seen with our customer segments. And that's kind of what we shoot for, especially in in the enterprise contract that we try to make.

Vita Koreneva (AlphaMille):
And I guess, in the very, very important question from companies that already have set up IT teams, why would they go with you, rather than try to build something in house?

Daniel Morozoff (VidroVR):
Yep. So I mean, I think the argument here is, is like, you can invest in trying to build this in house, but you effectively run a number of risks. One risk is the simplest one is that it might fail. The second one is that you need to hire, you know, Premier AI callin to build it, which is very difficult to do and very expensive to do. And it takes time, even if you do hire them. And you know, that takes time on the order of years. So your overall investment in this, you know, as a business will run at least $10 million, right over these years. And you will delay and risk failure. So the simplest answer is, it's hard to do. And it's already done. And you can get it for much more cheaply than you would if you tried to build it yourself. And the analogy here is like would you try to build your own car, even though you can go out and buy a transmission, a suspension, an engine, right, like very few people actually go out and build their own vehicles? Because it is challenging to build The system. And that's what we do we deliver a system that uses state of the art components that we either build in house or basically manufacture off of other focuses that basically deploy them either publicly, or we buy them if you need specific things we don't support out of the box. And then we basically deploy it. And that system stands up and it runs and it works. And we guarantee it provides you an SLA, etc. So the short answer is, you can try to build it. But obviously, that runs a number of risks that are business risks, that if you're willing to subsume for your business interests, that's one thing, but most of the folks that we deal with don't, and a lot of the companies that tried to build it failed and became our customers.

Vita Koreneva (AlphaMille):
Well, you also have a patented technology.

Daniel Morozoff (VidroVR):
Sure. I mean, we don't we're not in the business of being a patent troll. So our job, we have a patented technology. And the reason we have a patented technology is because it is something that we built, when we were still students, the university, patented it and licenses it to the company exclusively. So it gives us some level of defensibility of our ideas and our work. But we don't pursue other people who compete with us, we believe that our technology speaks for itself. And we compete on the merits of that technology, that's not your business, essentially, businesses to provide value to the companies the tools that you have.

Vita Koreneva (AlphaMille):
So a lot of people have concerns for how AI is developing, the speed of which is just boggling people's minds. And actually some of them are threatened by it. AI technologies seems to be developing very fast, and a lot of people are excited, but there are many who are concerned. Because you know, they're threatened by it, what is your take on AI technology and where it's going and the development of it in the next couple of years.

Daniel Morozoff (VidroVR):
So I'm definitely in the camp, believing that AI is something that is not a human replacement. Everything that I've seen to date doesn't replace human beings. All I mean, in the grand kind of holistic nature of what a human being can do in any role, what they do, what AI is very good at is very specialized tasks, where they're optimized in performance in those tasks. And this is true across the board from perceptual technologies, where, you know, AI, computer vision technologies are able to detect objects reliably now, at high degrees of performance and are reasonably robust. Similarly, in audio and other types of modalities, you see the same thing, text etc, where you start seeing failures are kind of at the tail ends of the distributions, where effectively there's bad representation on the training data is hard to generate, where, excuse me, where human beings Excel, and that adaptability is very hard to replicate in an autonomous or or a computer system. And, to my knowledge, very no one has actually solved that problem. Because of that, the way I view artificial intelligence is effectively an enabling technology, it is a tool that will revolutionize our world, but it will revolutionize our world not at you know, at the cost of human beings, but rather at improving their lives holistically. Now, does that mean across the board people will not have certain things that they do replaced by a machine or an algorithm? No, it they will be, but those people will be shuttled into another job or another role, where there will be an opportunity to have work and provide value in the economy where machines are not good. And holistically on average, as as a society or as a group, our output will increase and our efficiency will increase. Now, whether or not there are considerations and things to consider around like AI technologies applied to different scenarios, such as military and like ethical questions where these things do matter, and need to be considered, because those small augmentations do impact, like, fundamentally, like, I would say, fundamental decisions on like life and death. This is not just in the military, but like in medicine, or in space communications or in like, even autonomous vehicles, right, like I think everyone has seen that the promise of autonomous vehicles driving at level five anywhere in the world, is is a ways away, no matter what the marketing teams say, there are still no autonomous vehicles driving in New York City, right. And the news is, again, these tail distribution airy effects that lead to significant, you know, degradation in performance. But in the condition of a vehicle, that's a big deal, because if you crash, there's a likelihood either you die or someone else dies. And so the cost benefit analysis that happened in those scenarios is are we good enough to accept? You know, an average performance characteristic on a vehicle That's an ethical question that needs to be answered socially and societally. Again, I think those are challenges we need to kind of face head on because the world is changing in this direction. And we need to accept it and kind of figure out how to do it properly.

Vita Koreneva (AlphaMille):
Yes, thank you so much. What I was thinking you said, a keyword augment. And it sounds like it's very valuable for businesses to consider because essentially, what what your tool does, and what we see at alpha male, when we work with companies, financial companies, to help them with machine learning tools, implements certain, certain solutions that will save time and save costs, and etc. It's not necessary to replace people, analysts who review this materials and review or do research on investments, it's essentially help them correct, help them make the decisions faster, source information faster, and then analyze that structured data faster, so that they can actually get more done, but also really work in another area if you wanted to. So it's essentially maximizes the time it's not to replace the right.

Daniel Morozoff (VidroVR):
Yeah, so it basically gets you I would say, like, if you think about the funnel, it basically takes the top of the funnel and gets you all the way down to the bottom where you your work can be maximized against a larger base. And that base is that unstructured database that's very, or like, that's very hard to work with, and requires a lot of like, massaging to actually start getting meaningful signal out of. And so that part is where our system works and kind of provides value so that you can focus your efforts and energy on really extracting the meaning from signal, which machines are not good at.

Vita Koreneva (AlphaMille):
Right. So you, I'm glad to hear that humans will not be replaced.

Daniel Morozoff (VidroVR):
Not yet.

Vita Koreneva (AlphaMille):
Oh, boy. Okay. Okay, wonderful. So in terms of your company's terms, in terms of your drover, I understand that you're already expanding geographically. So you're not just a national company, is that correct?

Daniel Morozoff (VidroVR):
Yes. So we are starting to do business in various parts of the world, including kind of Latin America, Middle East, and kind of like we're slowly expanding. Because again, I think this speaks to the fact that everyone shared the same problem sets across the board. And those problem sets are consistent in various like industries, regions of the world. And so we're able to guess, convey the value of being able to tap into those resources or those unstructured data sources to really drive kind of value in these various sectors. And so it and it resonates with people across the board.

Vita Koreneva (AlphaMille):
Absolutely. Everybody wants to have an information advantage. Thank you so much for your time. Thank you. Appreciate it. Till next time. Bye