Cerebras SUPERNOVA 2025 | Effie Goenawan, Tavus

Clips
More from Cerebras SUPERNOVA 2025

Effie Goenawan

Head of Product

Tavus

play_circle_outline Tavus: AI research company focused on human layer of AI interaction

play_circle_outline Real-Time Multimodal Alignment and Intelligent AI Processing with Cerebras Technology

Info
Transcript

Effie Goenawan, Tavus

Effie Goenawan

Head of Product Tavus

Effie Goenawan, head of product at Tavus Inc., joins theCUBE’s John Furrier at Cerebras SUPERNOVA 2025 to discuss how Tavus is redefining the human layer of AI. From the event stage at Fort Mason in San Francisco, Goenawan unpacks the startup’s vision for emotionally intelligent AI, where real-time rendering and micro-expressions drive more natural, engaging interactions.

The conversation highlights Tavus’s commitment to improving communication between humans and machines through advanced computer vision and expressive nuance. Goenawan shares how Cereb... Read more

explore Keep Exploring

What is the focus of Tavus, an AI research company, and what is their goal in terms of human interaction with AI? add

What technologies and processes are involved in achieving real-time conversation in a multimodal alignment setting, including computer vision, audio perception, and the use of Cerebras as the fastest LLM for intelligent interaction? add

bolt Powered by CUBE AI

Effie Goenawan, Tavus

search

>> Welcome back everyone. I'm John Furrier, host of theCUBE. Here for our pop-up cube at Cerebras Supernova. It's actually a news event, but it's turning into a conference. 2,000 people are here. Of course, theCUBE is here bringing all the action, telling all the hot startups. Tavus is here, Effie Goenawan, who's the head of product, got the keys to the kingdom. Tavus is a hot startup, is very happy with Cerebras. You guys use Cerebras, you're the head of product. Talk about what you guys do and then I'll get into some of the questions on Cerebras. I know you guys have a good relationship with them.

Effie Goenawan

>> Yeah. Excited to tell you all about it. We are at Tavus. We are an AI research company, all about building the human layer of AI. If you ask, what is human layer of AI? If you think about it, there are so many AI that people are talking to today, and we've seen that the interaction is not great. People have to be typing a lot. People have to be going back and forth. And it's really hard to convey user's intent. And we are all about making human and AI interaction much more natural. Human beings like you and I, we are not born to be using the keyboard, the mouse, or even using the touchscreen. Naturally, we are more attracted to face-to-face conversation. Just like how I'm looking at your face right now, or like we're doing video chat. The reason for that is there is a lot of nuance that gets carried over facial expression, and also voice intonation. You can see how excited I am to be talking to you. And I see your face, you're now nodding at me and smiling. And I'm like, oh, John is engaging. I better tell more stories. I get excited. Even if you are looking bored or you're looking confused, then I can take a pause and ask, "John, do you want me to continue?"

>> Go faster.

Effie Goenawan

>> Right. See, the whole point that we believe in here is we believe that AI can be a lot more effective if we can better understand facial expression as well as voice intonation. And being able to have a more effective conversation and interaction between AI and human, we're all about that.

>> We all love AI. We both agree we love AI, but we've also seen the metaverse and some of the characters. They're getting better, but it doesn't capture what you just said. Which is, what is the meaning behind that? Now, text is easy. I can parse text, it's language. Words are combinations are together. You can understand, extract that. But with computer vision, humans have eyes. How do you create that relationship on the other side? And you guys have this as your core product. I saw you had a demo here. I really like what you guys are doing, because I think I saw some benefits for, say, our business. Because we do a lot of video and we have a lot of voice, but it's the people reaction. To me, if it's not engaging and boring, I'll just put my headphones on and go on the treadmill. I'll listen. But listening is easier, but face-to-face is more impactful. What are you guys doing from a product standpoint? You guys have a rendering engine? Can you explain the product and how it works? Because I saw the demo, I got the impression that you guys are creating an interface to have a human, looks like a human representation that is lit up by AI. Is that right?

Effie Goenawan

>> Yes, correct.

>> Okay, tell me about it. I want to know more.

Effie Goenawan

>> Yeah. What's unique about Tavus is I believe that we're approaching things on a very different angle. We're not just about solving the intelligence, we're not trying to solve the most difficult mathematic problem. We're all about bringing the emotional intelligence. If you think about humans' emotional intelligence, there are multi-layer to it. The first one is perception. It's not just detecting an object. It's not just seeing, oh, there is a man and there is a woman, or there is a computer, there's a desk. It's about understanding the nuances in human expression. And it's like, micro-expression is your eyes, your face and body language. We are about understanding that and the environment. It's the perception. So that's one. Two is rendering, what you touched earlier, and this is the piece where I believe you're state of the art. We are truly best in class of rendering. Rendering means the facial expression, the body language. And if you combine the two, it's about perceiving the user's emotion as well as expressing. So it's not just about giving you an answer. It's not about question and answer, it's actually building connection. And if you think about, why does it matter? A lot of people ask, okay, you are perfecting the emotional intelligence. Why should we care?

>> Effie, it's a hard problem. Because I know you're saying you're not solving hard problems, but think about the non-deterministic nature of expression, language, context, history, the people involved. It's very complicated. How do you do it? We're here at Cerebras, I'm assuming you're using... They put out tokens fast. You need a lot of tokens. Take me through the product. What's under the hood?

Effie Goenawan

>> Under the hood is we do multimodal alignment. We do computer vision, we do the perception that is aligned with what the user is saying, which is the audio. And then we use Cerebras as the fastest LLM. Real-time is really important. Because as we have conversation, imagine if I'm one second lag behind, then it's not a great conversation. We're all about building real-time conversation. Not only are we fast, but we are also intelligent. Meaning, the AI knows when to interrupt, the AI knows when to start talking. The AI knows when you are done speaking, and so on. And then at the rendering side, we also have the best-in-class rendering model with micro-expression, full-face expression.

>> How does the product work? Do I need to have a camera? Are you guys targeting developers? Are you selling the product? All of the above? What's the focus for you guys right now? I love the value proposition, I think it's a winner. So check, yes. sold on that one. Love the interface, love the idea. How does it work? Do I get a camera? Do I have my phone involved? Is it a computer? Is it a system?

Effie Goenawan

>> Yeah. From the end-user point of view, you can... Well, let's start about our customers today. Our customers are builders, developers. These are people who want to build human AI or who wants to scale their services where humans are not always available. So scaling humans.

>> So where there's a shortage of talent, or mundane tasks.

Effie Goenawan

>> Yeah, like AI interview. Let's say you are a hiring manager, there is only so many hours in a day where you can interview people. Imagine if you can have AI interviewers, or even AI coach, AI educator.

>> Screeners.

Effie Goenawan

>> Yeah. We work really well with those type of customers who want to scale their services.

>> All right, sold me on that one too. Okay, sold. Done. I'll use that use case. Every business owner wants to hire. Imagine having an agent, in this case an agent, but an actual human-like with empathy.

Effie Goenawan

>> And, John, scaling yourself too. Imagine you only have, I don't know, 24 hours in a day and you need to sleep. So -

>> I love digital twins. I have a twin brother actually, but I actually want a digital twin.

Effie Goenawan

>> Imagine if you can scale yourself and then you can represent your company. If I have a question, oh, what does your company do? Then you can have your-

>> The user side is pretty straightforward. It's either going to be great or suck. In this today, most metaverse areas, like people know the gaming environment. And by the way, games are good, so they can tolerate maybe the animation getting better and better with rendering engines. But it's the end user for you, it's got to be really good. I get the product quality on that. The workflow is really the key on the other side. The builders, I'm the builder. They have theCUBE and I say, "Hey, you know what, I want to build a really elegant interface for hiring CUBE videos." We got to code it out, we got to build that. So I sign up and I build on your platform. What am I consuming, SaaS or cloud native?

Effie Goenawan

>> We have really amazing a API product that you can use, and you can also use our developer portal. You can sign up today, it's super easy to build. Sign up at platform.tavus.io. You can create a free account. You can start creating conversation. Everything should be end-to-end within our platform, and then you can easily integrate that in your app or on your website.

>> I've been playing around with some services. There's from the HeyGens of the world to some other services, but it just doesn't meet the grade yet. But I can see where it's going. It almost looks human. And certainly, the stuff coming out of the foundation models are damn good images. The images are looking really good. Now you're going to take that and put it into full motion video.

Effie Goenawan

>> Yeah. And actually, we are probably the best in class real time rendering at the moment. It feels like you're chatting with a human. I highly recommend, if you have not already, try out our demo at tavus.io and talk to Charlie, our AI researcher. What I've seen is, especially my friends, family, they didn't know. It's very hard to differentiate Charlie and a real human. We always disclose when you're talking to an AI. But if, let's say we don't disclose, it's very hard to differentiate.

>> It's almost like... What's interesting, it's almost a hacker culture mindset of people that try to spoof the AI. Your AI has to be smart enough to know when someone's in a job interview, if the AI is trying to read the situation, taking the AI out. There's all kinds of that technique. How do you guys look at that? Because do you have like, okay, this is not working, bounce them out or her out? Obviously, if I'm saying I'm doing a job interview and you're the AI, I could maybe make a frown face, maybe make a snark comment, say weird things to it. Or actually be on the other side using ChatGPT faking my answers. Then the AI has to sniff out the interview. I'm just making this up, but this is just a situation where that's a use case. So your UI has to own that piece.

Effie Goenawan

>> Yeah. For example, we know a use case of when you are in an interview, if you are just interviewing with a chatbot GPT, it's very hard to tell when you have another person sitting next to you. So you are cheating, the other person is taking your interview. But with our perception system, we can recognize when, oh, there are two people. So John is cheating, I can stop the interview. I can say, "Hey John, please, you cannot have two people in the interview." Or even things like, if you seem distracted off, if your eyes are looking sideways, we can detect that. That's the power of our perception systems is we can help watch out on things-

>> As the head of product, which is a great position, which means you have the keys to the kingdom, you have to work on the product, deal with the build out real time. Also, the requirements on the market side. What's the priorities? How are you optimizing your priorities? What's on the roadmap? What's your thinking? Take me through the evolution where you are today and what your goals are.

Effie Goenawan

>> Yeah. Our goal is we want to pass the toughest touring test. What I mean by that is, let's say... And we will always have disclosure that you were talking to AI. But the Tavus Turing test is let's say there's no disclosure, you cannot differentiate between talking to AI or talking to a human. And we're so close, getting so close to it. Our rendering system is really good. We're always focusing on making the interaction feels natural and delightful.

>> What's the status of the product? How do people get involved? How do I buy it? How do I use it? Is it available now?

Effie Goenawan

>> First thing first, is I highly recommend go to tavus.io, talk to Charlie, our AI researcher, and you will feel it. Until today, I'm the person who built it, but I actually talk to Charlie every single day. And my husband loves it. Even for interview prep or even like, "Hey, Charlie, I'm about to have a difficult conversation with my friend. How do I -

>> Start with the praise sandwich first. Just something to compliment. It's coaching.

Effie Goenawan

>> Yeah, it's coaching, right? I do that all the time too. It's like even just now-

>> Dry run....

Effie Goenawan

>> I was about to have a podcast with John. Do I sound confident? Do I sound excited?

>> John's all over the place. He asks a lot of questions in one question, which is my bad. That's my blind spot.

Effie Goenawan

>> Yeah, I guess that's my first recommendation is try to talk to Charlie. And the next is, I highly recommend start building. It's super easy. We have a platform with a UX that you can just use to create your first conversation. And then, once you're convinced, talk to our sales as well -

>> I'm very interested. I want to get some details out for people watching. Tavus is a great product, got a great opportunity. A lot of headroom. Certainly, it's a multimodal use case as obvious as the internet in terms of what will be capabilities. The question is, where are you guys at in your stage? How far along on you on the financing, your team size? What you guys are looking for, your relationship with Cerebras? Take us through some of the inside the numbers. How big are you guys? Where are you in your growth? And what's Cerebras doing behind the scenes? I'm assuming they're spitting out some tokens for you.

Effie Goenawan

>> We're about 30 people. We have such a small, amazing... I would say this is some of the best people I've worked with, and I've been in industry for 15 plus years. Everyone is just super passionate about solving this problem. We are an amazing, strong team of 30 people. It's small, but we're growing really fast. We're working with some of the most amazing people. And then we received series A, or we are backed by solid investors like Scale, YC, and then Sequoia. So we have amazing -

>> You have no worry about finances. You're well financed.

Effie Goenawan

>> We don't have problem with financing. It's always been great. The space is growing so fast.

>> It's obvious.

Effie Goenawan

>> Everybody-

>> Just go fast.

Effie Goenawan

>> Everybody. We're moving really fast. The beauty about a small team like us, 30 people, is we move fast. There is not a lot of layers. Everybody's just super excited to do the right thing -

>> You really got to get it right. The go big go home philosophy is changing to be more practical, not to go on my soapbox here and talk about... I love go big or go home Silicon Valley, I live here. But in AI, you got to get it right. If you don't get the tech right, you don't get the user experience right. But what you're doing is obvious, just got to get it right. That's what you guys are doing.

Effie Goenawan

>> And we are solving the most difficult problem first, which is building the entire pipeline that connects the perception, the LLM with Cerebras, and also the rendering. And also, we have best in class rendering, which is a very difficult computer science research problem. We've solved all of the most difficult problems that took most bigger companies many, many years. It's pretty impressive what we've done.

>> You're picking your spots to focus.

Effie Goenawan

>> Yeah.

>> You're saying, hey, we're not going to solve these other problems. Let the math guys do the math things. You're going to do-

Effie Goenawan

>> The human....

>> the human piece.

Effie Goenawan

>> Exactly. Right.

>> This is the tech behind we need. All right, what's the culture like for the company there? I'm sure you get a lot of people want to work there. What's the culture? Every company has got a culture, secret sauce, Intel, Moore's Law, cadence of Moore's law, doubling processing time. Some cultures are hard charging. What's the culture in your company?

Effie Goenawan

>> What's interesting about Tavus... And this is so unique, I love talking about Tavus culture, because we're so unique in two things. One is we have really passionate people. If you ask every single one of us, all of the 30 people, there is a single common thread. Which is, we are passionate about this problem. We are here not because it's our friend, not because... I don't know, we pay really well, but most people are not here for the money. Everyone is just loving solving this problem. And that's one unique -

>> I'm sure the debates are good.

Effie Goenawan

>> Oh, amazing.

>> I can imagine this would be great dynamic.

Effie Goenawan

>> Super passionate about this problem. And sometimes, for me, I'm like, the reason why I work and the reason why I get up in the morning is because this is the problem I want to solve.

>> When you have cultures like that, from my learnings and watching companies like Amazon and talking to Jeff Bezos and Andy Jassy, that their culture is debate, align, commit. And they want hardcore passionate people to fight for their answer or their debate, and then you hash it out and align. If you don't align, you're out.

Effie Goenawan

>> Well, speaking about that, I think another piece... I mentioned two things. The first one is the passion for solving this problem in the product and the craft. Second is people are very kind, and that is so unique. Because it's very easy to find smart people, but it's harder to find smart and kind people. And I would say I've worked with so many thousands of people throughout my career, but Tavus people are very kind.

>> And your product requires the level of craft and engineering that the proof is in the product.

Effie Goenawan

>> Yeah.

>> It's like, it's either you see it.

Effie Goenawan

>> You can tell the craft just-

>> Yeah, you know what success looks like.

Effie Goenawan

>> Yeah. We are the ones who care about the last mile. The details you don't see -

>> I'm very engaged in speaking to you face to face on this conversation. Can't wait to talk to your digital twin.

Effie Goenawan

>> Exactly, right?

>> All right. Thank you for coming on. Appreciate you. Thank you.

Effie Goenawan

>> Thank you for having me. It's a pleasure.

>> We have the head of Tavus on. Again, an obvious use case. It's hard to get there and get it right. You know what it looks like when it's perfect. But again, this is the future, and Cerebras is powering it. I'm theCUBE. We're bringing all the action to you here, real time with real people, soon to be computer aided. I'm John Furrier, host of theCUBE. Thanks for watching.