AI-Ready Data with illumex’s Inna Tokarev Sela

This is Textron tv. Hey everyone, welcome back here to another Techron TV interview. I have a new company and a first time person here on our show to introduce you to, and it's about ai.

What a great thing to talk about here. Before we get into our holiday breaks, I wanna introduce you to Ina Tucker. Ina is the founder and CEO of a company called lumix.

And, uh, it's a gen AI data platform for enterprises. She'll tell us more about that in a second, but let's welcome Ina, thank you for coming on the show. It's a pleasure to have here.

Happy to join you today. Ah, it's our pleasure to have you on. So, ina I always like to let people get an idea.

They're talking to, as I mentioned, you are the founder and CEO of Lumex, but you know, as they say everybody, to start a company, it's like, it's almost like having a child, right? You, you gotta put a lot into it. It, there's a lot that goes into it.

You, it, it, it's a big part of your life, so you've gotta be very passionate to do it very. You've gotta have a, you know, really feel strongly that in some way what you're doing is gonna change the world, at least in a little way or make the world better. Talk to us about your background and what you do to Founder Lumex.

So I was passionate about how enterprises can really manage their business with this deep core understanding of every process and every workflow and how low the data service was around his organization, despite the fact that companies invest so much in data platforms and analytics. Gartner says only 20% of decision making in enterprises actually made based on data. And I started my career at SAP working with Walmart specific builders, or the word, you know, huge enterprises, source for data analytics and build amazing things.

But this friction about how you manage data, how you clean it, how you model it, and how you eventually serve it to, to the customers is, is a very lengthy process. And this is why I started Elmax to, to remove this friction between data and consumers. Excellent, excellent.

Um, 25 years in, in data, but not in ai, right? It, let's talk for a minute. Do you think you would've started the Lumex if we didn't have this whole gen AI revolution going on?

Uh, it's excellent question. So depends how you define AI really. So I was a graph junkies from my first degree, you know, coding it in MAT labs.

Uh, so I absolutely love the technology and generative AI has different manifestations and instances in this current form. It started 2017 and I had the pleasure to, to a workforce generative AI models both in video analytics space at SAP and then in business analytics at Sisense. So to me, what happened in the recent years, uh, different factors which brought generative AI to the front of the stage and mainly due to consumer interest.

So chat, GPT exploded in 2022 and suddenly business executives were excited about this technology as well. It's first time since internet, that business actually comes to data and technical people and asking for something, right? Mm-hmm.

It is. And anyways, that, that isn't important. I, you know, one of the things I tell people about the AI thing is that, you know, we've had technology innovation over the last 25, 30 years, but we really haven't seen this kind of society-wide upheaval or involvement since the internet.

This is more like when the internet first became commercial than it is, let's say like when the cloud first came out or something like that. The cloud, everyone in technology got very excited about the cloud, but in society in general, you know, it was less understood and it made things easier. Ai you could talk to people on the, I went to the veterinarian yesterday to my dog needs surgery, and the veterinarian was telling me how he's a big check GPT-4 O user and it, it's helping him, right?

And he's not a technical person, but there you have it. Um, let's talk about the founding of ium. Tell us how, how Ium X came to be.

Yeah, so already few years back, uh, I understood that the generative AI is coming. And I also saw that in enterprises we are still not, uh, ready for, you know, to accelerate our data access on the scale for generative AI proverb, self-service to, you know, to every employee because data is frankly speaking not good enough, right? And our understanding and our management of data is not scaled enough.

So that's why I started. Ax, ax helps enterprises, especially data intensive and highly regulated ones, first to bring the data to AI already states. Second, being able to govern it because I do not believe in generat AI without governance.

Absolutely not. And the third one, to provide hallucination free and explainable answers to the end users. So we're taking this full stack of general AI productization for enterprises to make sure that the whole process is monitoring guide drilled and trust sourcing.

Absolutely. Let, let me see if I can help kinda set the table here with what the problem is. You know, we all have heard stories about how, how, you know, some of the big gen AI models like Aachi, GPT or, or, or Claude or, or Gemini or what have you, have, they just basically suck data right outta the internet, basically, right?

Whether you knew it or not, especially like for us, right? We're a publisher, we publish a lot of information, a lot of content. Well, our content was used for sure by some of these models without our permission, without our compensation with us, you know, not being, but it was used.

So I think on one hand people were concerned about, hey, these people are building a model on my data and, and without my permission or what have you. But then secondly, so let's call that unintentional, right? I, I didn't know they were doing it, they didn't.

But then you have intentional data users where I wanted use gen AI to do something in my company and a general data set, one of these huge LLMs isn't enough. I need, I want a data set of my data, but it's gotta be the right data. It can't just be everything in anything.

And, and I'm gonna try to figure out what's right or wrong. I wanted this particular data to be the, the data set that the model uses, I think blitz. And so that's intentional, right?

I intend this data to be used. I think that's right now where we have a huge hole in the model and that people don't know how, yes, they could say, well, I'm just gonna take it in a vector database, right? That was what I heard a year ago.

Oh, no problem. Just put it in a vector database and it'll suck it right in there. Now we're hearing, well you got graph, you got this, this rag.

There's a lot of ways of getting the data into the model, but it really comes down to you have to kind of pretreat that data. You've gotta pre, you gotta make sure it's the right data. And it sounds to me that that's what a lumex is really about, right?

Trying to make sure that the data you intentionally wanna put into a gen AI model is the right data in the right format with the right tags and so forth so that your, the, the, the whatever AI you choose to use is getting that data the way you want it to, to be had there. Exactly. Correct.

To, to your point, you know, to about vector databases or graph databases. You know, a decade ago, uh, companies thought that just moving that data to cloud and it somehow will, will solve the problem. And we have the incarnation of the same approach with whatever new technology it is for generative ai.

So it doesn't solve the problem by itself. It's cost prohibitive to me to actually move your data to ai. You need to bring AI to where your data is, right?

So to, to mitigate costs and all of that. Uh, so I mentioned a few things. Uh, your data, right?

Your language, your content, we call it semantics, right? And generative lingo. Understanding your company semantics is imperative for generative AI to avoid hallucinations.

And the second is context. The context of your question, context of your definitions on your data context of the usage is very, very important to, uh, for, for prompting as well to get correct a answers. So both content and context have to be built on top of your data, across your landscape.

So your data lakes, your warehouses, your supply chain systems, all of them needs, need to build into context and reasoning for generative AI to be used. The thing is, we have this teams of data scientists, many of them, right? Building this context and reasoning and tools.

So frameworks, which you call drug graph, rock and other approaches, but those people are coming from scientific backgrounds. They're not business experts. So again, we have this reincarnation of data modeling, which we already had in bi and we already created like many versions of truth and laws and translation business analytics.

So why we're saying that it's a good approach for generative ai and now it goes on scale, right? So if you implemented mistakes in your context and reasoning across your data, many of your users are going to expose to hallucinations. So all of those challenges I max solve automatically.

What we do, we actually approach a data sources by only touching metadata. We do not touch values because we know that the risk management is very important for generative ai. So we touch metadata out of this metadata.

We automatically build context and reasoning because our platforms are already trained on domain specific ontologies. So we are, uh, domain specific. Our platform is not trained on internet.

And when we build this context and reasoning for organizations, we also provide workflows for domain experts to be able to review and certify. So I believe in human in generative AI loop. I do not believe in taking human out of generative ai.

You do not necessarily need to review everything, but you have to have ability to review everything. And, and for sure when we pick up all those conflicts and definitions that happen all the time, right? No organization is defined churn in the same way.

Different departments define it differently. And here we are surfacing all those conflicts. So those conflicts might be addressed by humans, but you do not necessarily want humans to build this context and reasoning for you.

Moreover, 80% of tokens as of today are spent off customization of fine tuning this off the shelf models on your data. So it's also super expensive to do it otherwise Got it in. If you don't mind, walk me through, let's say I'm an organization and I'm thinking about using alumax, right?

To, to make my data AI ready. Walk me through the, the sequence of events. What, what, what do I do?

How do I engage it? What do I do at each step of the way and wind up with AI ready data, you know, and my gen ai, as you said, bringing the gen AI to me instead of bringing my data to the gen ai. Amazing.

So majority of companies have AI strategy and they have use cases, right? So imagine you bank, let's say a use case is customer segmentation. So you'll identify which data do you need to, to ask questions about customer segmentation, customer tiers, customer preferences, all of that.

Let's say it's CRM database and it might be some, you know, operational system. Then we'll contact Elam Max on Elam Max ai, get, get our sales people impressing you with lovely demo our system. You spoken about babies and patient, I am so passionate about mm-hmm.

Elmax experience. And this, this really works to, to our benefits. So without customers or prospects with our demo.

And the next step would be really defining this scope of, of the project and identifies those systems. Creating grid only access to your metadata and few days later, you know, giving the paperwork is, is handled on the background. They can already play with the system and uh, they can onboard customers faster that we clear the paperwork and this is how it should be with generator ai.

Yeah. Yeah. It should, it should almost be like magical, right?

It just happens. It happens as far they can check it. I'm all, all transparency certain.

Let me ask you a question. Had you know customers who, who's the target customer here? Is it large enterprises with lots of data though?

Just because, you know, every, every organization today has lots of data. You don't have to be a large enterprise to have lots of data, but who, who is the customer? Is it larger companies?

Medium, all companies? Our ideal customers, uh, in, in this intersection of being data sensitive and highly regulated, uh, companies, which means AC accuracy is important, governance is important, risk management is important. And TCO total cost of ownership is important.

So financial services, pharma insurance, our targeted, um, industries, uh, but our solution is industry agnostic. Really Fantastic. ai, right?

Alright. Is the website. Yes, indeed.

You know, we're about outta time. I want to thank you for coming on and, and telling us a, the Lumix story and a little bit of your story. I wish Lumix lots of success for people out there who are now realizing that it's not just getting the data to the ai, it's bringing the AI to you and making your data ready for it.

Lumix is a company you might want to check out. We're gonna take a break here on Textron tv. We'll be back in a moment.