theCUBE + NYSE Wired present the East Coast AI Leaders Executive Series | Raj Rao, Lilt

Clips
More from theCUBE + NYSE Wired present the East Coast AI Leaders Executive Series

Rajesh Rao

COO & CFO

Lilt

play_circle_outline Lilt's background and vision for providing high-quality translation services

play_circle_outline The need for domain-specific modeling in translation to ensure accuracy

play_circle_outline Enhancing Enterprise Communication: The Impact of Accurate Translation in Regulated Industries

play_circle_outline The challenges and investments in scaling language translation services globally

Info
Transcript

Raj Rao, Lilt

Rajesh Rao

COO & CFO Lilt

TheCUBE is discussing Lilt's mission to provide high-quality translations for enterprises, focusing on AI and human-in-the-loop collaboration. Lilt customizes models for accuracy and efficiency in different languages and industries. Raj mentions challenges and growth opportunities in the language translation market, emphasizing cost reduction and real-time translation. With over $100 million in funding, Lilt's priorities include marketing, product enhancement, and AI infrastructure development. Language specialization in AI models, accurate translation in reg... Read more

explore Keep Exploring

What is the history and vision behind the company founded by two Google Translate folks with PhDs in natural language translation, focusing on providing high-quality translation services using a combination of AI and human input? add

What type of enterprises is the speaker suggesting would benefit most from their translation services? add

What industries require accurate and precise translation due to the high cost of mistakes, and why would they choose to work with a company like Lilt? add

What has Lilt's mission been in relation to the expensive and constraining factors of the localization market? add

bolt Powered by CUBE AI

Raj Rao, Lilt

search

>> Welcome back everyone to theCUBE's coverage here on the East Coast studio for theCUBE here on the balcony overlooking the New York Stock Exchange on the floor behind us, opening bell. We're into day two wall-to-wall coverage on Wall Street, theCUBE connecting Silicon Valley and Wall Street in our new studio locations. Raj is here. He's the COO and CFO of Lilt. Great to have you. Good to see you last night at the dinner reception of the founders the founders of all founders meeting together.

>> Absolutely.

>> Great to have you. All the AI leaders were there. Great to see you.

>> Absolutely, yeah. It was a phenomenal event. Thanks for making it happen.

>> Thanks for coming on. So you're running the trains over there, making the trains run on time, making sure they don't run out of cash. That's a good thing, right?

>> Yes.

>> That's good. Being the CFO, you got to protect the cash. I always say the secret to startups is don't run out of money.

>> Yeah.

>> So Raj, you guys do really cool translation language models. Obviously. Language is not just English. And then you also have other languages. Machines talk to machines. Processes are its own language, we always say on theCUBE. Talk about the company, where you guys are at right now, size, formation stage. What's the stats?

>> Perfect. Okay, so just to give you a little bit of the history of the company and so you get a sense of who we are, the company founded by two Google Translate folks who met there. Both have PhDs in natural language translation. Our CEO is now one of the co-founders. And the vision was really about getting access to information to the entire world no matter what language they speak. That was the real vision. Initially, the perspective was we do it purely through AI, and this was AI before ChatGPT. It was the much more visionary thinking back then. But over time we realized that the real opportunity is with enterprises where the cost of making a mistake is extremely high and you need to offer translation of the highest quality. So what I think our real secret sauce is to bring AI and the human in the loop to deliver extremely high quality at a much lower cost than average. So our classic customer is Intel, for example, large enterprises. Imagine they have marketing content, product content, legal content. All that needs to get translated into different languages. You need to have the right level of quality to deliver that. So we would be the right solution provider for someone like Intel, and they're one of our key customers.

>> Translation comes up a lot with theCUBE because we'd love to have our Cube interviews translated into foreign languages. But it's small. We're a small language model, but I can see the value of continuity in the business as a big factor. How often does that come up? Because most of the conversation is around governance, but governance is also a part of making sure you don't get the language wrong when you have processes. Talk about that factor.

>> So if you think about regulated industries like financial services or life sciences where you need to make sure that certain content, like your prescription specifications, your compliance documentation has to be really translated into the local languages and it has to be accurate. There's no room for error and so the cost of a mistake is extremely high. A good example of that, John, is National Weather Service. So we are the official translation engine for them. They have the high risk of, if there's a mistake in how the warnings are communicated and the speed at which the warnings are communicated, there's lives at stake and so there is no room for error. And I think that's why they'll work with someone like Lilt.

>> And the business model of subscription?

>> On the enterprise side, we have the software platform, so there's a subscription model for that. But the key thing is that on top of that we have human in the loop. So we actually have a base of over 6,000 translators who actually work on the platform. They're using our AI enabled solution to make them much more efficient. And so, actually, there is a software plus a usage based component on top of it.

>> How much is the language English versus foreign language out there in terms of scope, the magnitude of the delta between English and foreign language and character. Obviously, the internet was first originally ASCII, which was English based, then it became Unicode based. Language, these models are all trained on mostly English. What's the scope of that?

>> Great question. So I'm just stepping way back. If you think about it, roughly about 400 million in this world where English is a native language, if you include people for whom English is not the native language, that's about a billion people. That leaves about five billion people who don't speak English.

>> right.

>> And as you exactly said, the models, the large language models, roughly 95% of the training data is in English. So what's really important is that you have domain-specific modeling and that's where we come in. We do hyper-customization, so we have 17,000. We have our own foundational models for translation and we hyper-customize it for the customer. And it's the language pair, it's their content, it's trained on their specific content, and within that there's within the domain specialization. You probably have read this a lot, the general models generally do fine, but when you actually get them deployed for a specific application, they can very quickly go off the rails. And that's where specialization is super important.

>> It's interesting, people who are in the know in AI who have been around for a while, even anyone that knows linguistics, word combinations matter, especially in technology where jargon is important.

>> Yes.

>> Knowing the nuances of combination of words is super important. This isn't just hit a button and translate a transcript. That's not what you're doing. There's deeper than that.

>> Yeah. John, you're spot on. If you think about a brand, they have their own, there's a whole tone of how you want to communicate to your brand message and it has to be captured. The essence of that is captured and you can immediately tell if it's a machine translated or a AI translated one. So I think the domain specification, we spent a lot of our specialization in training our models to capture that domain specific knowledge. Everything from terminology to the tone of voice to historical phrases that they want to be kept exactly the same. That's where the in-depth knowledge comes in.

>> Okay, so you have me sold on this. I love the product. Obviously, we do CUBE interviews, we're in English. We have a feature where we have our own little language model with a CUBE AI and we have a feature where you can transcribe the transcript and it just flips it into the language. And again, it doesn't translate. We spot check it. I can't really check how well it translates to Japanese, for instance, or Korean. But when I show people, they're polite when they say, "It sucks. It's not good." And so this is where I like the opportunity because I would like to buy the service. How do I engage? Pretend I'm a customer, what do I do? Do I connect via an API or is it a service? Do I move my language model over? What's the process? Take me through that because if I'm a customer, I want to get involved.

>> Absolutely. So effectively I think if you look at our product, this is important again in the enterprise context. You're a customer, you probably have content sitting on S3 somewhere and then your Google drives. It could be even a PowerPoint presentation and you're just like, hey, look, figure out, how do you get all this content from my content management systems and translate it and publish it back to me? You don't want to be emailing content. So the orchestration layer really that we have built hooks into your content management systems who grab the content, take it through our translation methodology, bring it back and publish it back. This is another differentiating factor that we have, which is the seamless orchestration. And now you might insist, no, actually, I'm going to bring my own model because .

>> We had our own , it's great.

>> We actually allow model interchangeability so you can bring your own model and hook into it. So I think this took a lot of investment to get to this place and that really enables us to serve customers with wider variety of needs.

>> I'm sure the founders at Google saw a lot of the scale challenges too. And just from a business model standpoint, localization has been around for a while. Every app has a localization plan. It usually lags the initial feature, but it's expensive too.

>> Yeah.

>> Now, you multiply that expense by the language models as training and then inferenced. Take us through that process of degree of difficulty to how fast you guys turned that around. Take us through that.

>> Absolutely. So to your point about how big this space is, localization is a $30 billion market. That's massive. And I think, as you said, it is expensive and it's been a constraining factor why more content is not getting translated and accessible, and that has been Lilt's mission. We have really cut through those cost barriers by using AI. Now, I think in terms of the degree of difficulty, as I said, there is a hyper-customization that the models had to be done. At the same time, we have translators across every single time zone across the world. They have to be in real time running the AI models. And the other really extremely valuable thing about us is the models are continuously learning. So as the human in the loop data is coming in, the models readjust over time. So the speed at which the models react, the infrastructure that we had to build to allow that to happen globally 24/7, I think that's been an engineering feat and that's why we made a lot of the investments.

>> Raj, talk about the costs involved, because one of the big things that I've seen in AI is the CapEx on build-out, as we see that with GPUs and whatnot. Operational costs also kick in. What's the financing plan? What's your investment thesis around capital? How much have you raised? What do you guys need? How's that playing out in the economics?

>> Yeah. I will not get into too much of the specifics, John, but I can tell you I think we have gone well past product-market fit. I think we now have a very established set of several large enterprise customers. Our biggest CapEx needs now is about distribution. So the investments we're now making is to build up our sales team, marketing, adding more channels. To your point, the AI models, the technology's going to continue to evolve and we'll continue to make the investments there. We have some capital-raised needs. We are backed with some pretty strong capital.

>> How much have you got raised so far?

>> Over 100 million so far.

>> dry powder. So you're ramping revenue?

>> We are ramping revenue, yes.

>> How's that going?

>> It's pretty good. I think '23, we came off a pretty strong year with 40% growth. This year in '24, we are likely to end with 60% growth top line, and the momentum looks pretty strong.

>> What's the vibe of the company? You guys wake up every day, you guys are obviously language experts. When you guys show up to work every day, what's the DNA of the culture?

>> So John, let me tell you what gets me excited and I think what gets most of our associates excited is, if you just think the promise of AI, you think about a tutor for every child. That is what Khan Academy is doing. We are already seeing some of that. A personalized doctor or lawyer, that's the promise. And I think the reality is possible, but the challenge is that most of the models that are built right now are in English. So being able to enable globalization of the power AI, I think that's super exciting. And by the way, in addition to all the translation stuff I talked about, we are now working with the large language model builders to enable to make those models multilingual. And we are perfectly suited because we have the AI expertise, we have the multilingual expertise, and the human in the loop expertise. All three are required for this.

>> It's a global economy. You got to have this in real time.

>> Absolutely.

>> And Generative is not pre-programmed. You got to be ready for everything.

>> 100%. And I think you mentioned Generative, John. I think the other thing that we're now branching and expanding our product base is now, not only can you translate, but also if you'd say, "Look, I want to write a blog in German, but it has to be related to theCUBE nomenclature, et cetera," we will just generate the blog.

>> Well, we talked about Kubernetes, you better get that right. These are the translation nuances. I mean, we've seen it many times. A lot of the dictionaries, old school methods don't really apply because there's always new language coming in. Jargon is a great example. Jargon really can help AI better because, if you can put context around Jargon, you can get so much more extraction out of that data. And this is, again, not well understood, but vector embeds and technology around neural networks, it is a brain and it's a human brain-like simulation of technology. So you got to have the context cues built into the retrieval.

>> Absolutely. And so I think, if you just think about the whole genesis of a large language models with the decoder, encoder, all was built off the translation problem statement. That's where Google built it, and I think that's really been the basis of that.

>> I think our traffic would go through the roof if I could just have an easy button for translating siliconangle.com, which is a great site. We do all the top reporting there, theCUBE Research, but it's so difficult to translate. My choices are hiring a firm to get it right, which is still not even a guarantee. So I got to have a mechanism to scale, and if I could stand up websites with low code or no code, I can have multiple versions of Silicon Angle up and running.

>> And John, to be honest, you are the perfect customer for us and we should absolutely talk.

>> This is a sales call. I'm in.

>> Yes.

>> Can I get a good discount, CUBE discount? .

>> .

>> In all seriousness, I think this is really going to be, I think we're living in a world where real-time but parallel innovation is happening.

>> Absolutely.

>> And this is where I think language fits beautifully into this notion of it's not like a slow roll waterfall methodology of rolling product out. You've got to roll out product at the same time. Globally, has always been a huge challenge, huge challenge. Whether it's an app translation or model integration, and, again, more models are coming out. Hugging Face, every day that leaderboard is bigger and better and with more models. And the models have a power law. You have the big models and you've got specialty models, and you've got specialty models that have proprietary data. I don't want to protect that data. I don't honestly want to have that data mistranslated. So even in enterprises, I can see them loving this.

>> Absolutely. And I think this is, frankly, one of our selling points. If you decide to use a large language model, I think there's two problems. One is it is not domain specific and not trained on your data. And B, you're sending your data to whoever the language provider says. With us, actually, it's contained within the domain. It's protected. And the most powerful thing is we are actually training it on your domain-specific knowledge.

>> What about transcription? That's another area that's come up a lot. We use transcribed systems for our videos in real time in the cloud. I want to go to transcription. That's also an opportunity.

>> We have focused predominantly on the multilingual aspect of it, but if you think about the multimodal nature of the data, we are seeing an explosion in terms of the different forms of data that's coming in or the media that's coming in. It used to be primarily text. Now we do voice.

>> Computer vision?

>> We have video, and all of that, and lip-syncing the whole thing. And that part of the business is growing really fast.

>> It's interesting. You think about computer vision for instance, you can translate what's on vision into describing what's in the picture. That's still language. Video we're talking, that has language, but it's video.

>> Yes.

>> That can be translated.

>> Yep.

>> That can then move. So there's so many vectors of opportunity for you.

>> Yes. I think our challenge would be to making sure we stay focused on the key vectors of growth and just do extremely well in that space.

>> And you have a problem in the sense, everyone's a customer.

>> Technically, yes.

>> You have to learn really to prioritize. What are your priorities now as you've got the go-to-market going, you got product-market fit, you got a lot of capital, you guys are well-funded. What's the priorities?

>> I think, as I said, the priority, number one, is really on the marketing and growth side, so investing in building our brand. I think we are seeing opportunities, as I said, in multimedia. So enhancing our platform to be able to handle various types of media. And then I think we'll have to continue to invest in the core AI infrastructure space because that is going to change dramatically over time and we'll always want to be one step ahead.

>> Well, Raj, it's great to have you on as someone who's got the keys to the kingdom as CFO, but also running the operations. Got great founder Pedigree at Google, got a great wave ahead of you. You're in the AI Leaders Network now here at theCUBE and NYSE Wired. You excited? You got to be excited.

>> Yeah. I mean, this is a great platform for us to talk about the company, and I think I strongly believe that all the excitement of AI making that accessible for the entire world using the local languages, I think is a really exciting prospect.

>> It's going to have impact. You mentioned education, Khan Academy. I'm a big fan of Khan Academy. Obviously. theCUBE videos, we're free content. This is a democratization opportunity...

>> 100%....

>> to change the world in the sense of truly Steve Jobs had that famous, "We want to change the world." I mean, this is happening.

>> Yes, yes. And I think we feel almost lucky to be in the center of it to be able to make it happen.

>> Put a plug in for the company for the last minute we have left. What are you guys looking to do? I'm sure investors are probably climbing down on your door, looking to hire. You've got goals. Put a plug in for the company and give the pitch.

>> A couple of pitches. One is, if you are a company with global aspirations looking to expand into different markets, you will need to think about globalization and localization much more methodically. And you need a partner who really knows this, who's AI forward, and loves the company. So come talk to us, visit our website, reach out to me. That's from a revenue perspective. Employees, we are always hiring for top quality talent. If you're passionate about democratization of content and AI for the broad masses, Lilt's the perfect place to be. And then, as a CFO, I'll never stop taking calls from investors. I would love to talk to anyone.

>> Valuation's going up. Get in while it's hot. Also, talk about the people in these countries. I can see you guys maybe having an ambassador program. Representation of all languages is the goal, obviously. How do I get involved if I'm watching in Japan or Korea or I want to be involved on that side? What are you looking for there?

>> I think we have a ever-expanding need for people with expertise in different domains as well as the local languages. So reach out to us on LinkedIn. We have a whole team on talent and development on those guys. We'll connect you with those guys to bring you in the loop.

>> Raj, great to have you on. Linguistics is hot. AI is all about words and combination of words. Translating them to the masses is really going to be the next wave. It's very not talked about, but this is the key to getting real-time product out into the market to a global audience. I mean, everybody wins. Check them out. They've got a great product. I'm John Furrier, host of theCUBE. Of course, we're doing our job to get things out, in this case, English here in the New York Stock Exchange at the center of all the action on Wall Street. We've got a studio in Palo Alto in the heart of Silicon Valley and here in the heart of New York City and the financial markets here, the New York Stock Exchange, of course. theCUBE's always covering the events, getting all that data, sharing that on our site, on YouTube, and, of course, soon to be multi-language partnering up. We'll partner up with these guys, for sure. Definitely want to be a customer. Thanks for coming on. I appreciate you.

>> Appreciate the opportunity.

>> Okay,

>> Thanks, John.

>> We are here at the East Coast center of the action, one of our super nodes, as we call it. It's a point of presence for our network. Join the community, be part of it, share the content. Let us know how we're doing. I'm John Furrier, host of theCUBE. Thanks for watching.