Rajan: AI is reshaping businesses and ignoring it could leave your industry disrupted. I'm Rajan and this is Pivotal Clarity. We talk to those building or using AI founders and engineers with real world experience. Our aim is to cut through the hype and see where AI is truly making an impact. If you're a business or following tech trends, these conversations offer, clearer insights than most of the press. Let's get into today's episode. Welcome to Pivotal Clarity AI podcast. I'm talking with Jeff Lu, founder of Akul, a company that uses AI to make video ads. Jeff, welcome to the podcast.

‍

Jiajun: Yeah, thank you for inviting me to the podcast. I'm happy to be here, to share some of the shorts and so on.

‍

Rajan: So Jeff, before we dive into the tech, tell us a little bit about your, pivotal moment from your childhood that sparked your interest in technology and computer science.

‍

Jiajun: So while I was at middle school, I'm really interested in reading all the books, and I stay, holding in the library and the bookstores, reading lots of books about Phoenix. So just very curious about the origins of the world and origins of their life and so on. That's the start of my interest in science. And then as time moved by, my interest in physics involves the growth. And during my undergrad I, picked, computer science and I picked computer graphics because in computer graphics there are lots of physics over there. They do huge amount of physics simulations and these kind of things, basically building a virtual world, a mirror of the real world. And so we are, building a world ourselves. So that's how I got started and then get into AI, which is building intelligence. So really interested in how the world starts, how intelligence starts. They're building this stuff.

‍

Rajan: Yeah, it looks like a lot of the folks in AI, they get started from some deep interest in physics.

‍

Jiajun: Yeah. Because physics is really the origins of the world and origins about life and all these things. Right. Physics is studying the fundamental things. And for the folks in AI, I think, it's same, it's studying the fundamentals of intelligence and the world of Phoenix is moving very slow now. So it's definitely a lot of people are moving to AI and it's a, fast moving space and a lot of connected places.

‍

Rajan: So you worked at big tech companies, you are now running your own AI startup and you've had a lot of years of experience in the industry. Based on all of this, what would you tell your first day at Google now, given what you know?

‍

Jiajun: So my first day at Google that's a while ago, four years ago at least. So I think that was during COVID So I onboarded Google during COVID remote at home and they already shipped me a laptop so I just use the laptop to onboard. But I think at that time I don't have their official Google account yet. So I just set up the Nugler account and get access to different things and learning different things know from home. So I think Covid is a different world. So definitely very exciting because in Google you can access so many interesting projects, you can read so many interesting stuff and Google build their whole own different architectures, many things to learn as well. So definitely a very exciting experience in Google. And the first day is mailing onboarding.

‍

Rajan: Is there anything from your startup experience that you will take back to Google?

‍

Jiajun: Not really but definitely in Google Cloud we are working on video processing and currently we're mainly working on video generation side still we have the thought of the cloud and we always have the cloud in the mind so that's something that we learned which is also different because in a generic AI space lots of startups is doing consumer stuff. But my previous experience in Google and so on are all like enterprise and the cloud based. So we are still more interested in the enterprise B2B and cloud business.

‍

Rajan: So Akul makes ads right? And you've made m millions of ads so far. How does one go about it from start to finish? What part do people do and what part does AI does?

‍

Jiajun: We help business to make videos, high quality videos and a lot of these videos are being used in ads and for now it's still pretty difficult if you just generate the whole ads from scratch. You still need huge amount of human involvement. So what we do is we do several things. First is we help them to personalize ads as well as translate ads into different languages and so on. So there need to be a video templates first and then we work on top of the video templates and make editing to the video to create variations of the video with high quality. And that's a huge part of how the AI video ads works. Some others we can create avatar intro ads and so on but usually for larger brands, larger brands they want some more commercial quality ads which is mainly creative variations. And we help them to localize their person to different countries geometries and then localize their language to different countries and so on. So these are the biggest benefits that we provide and actually Just two weeks ago, together with Qatar Airways, we launched a personalized global ads campaign that enables users to put themselves into a video ad that brings you travel over the world together with QOTA Airways. So that's one of the examples of the ads that we did. We do lot of different ads and we even help many brands and so on to create virtual influencers and virtual ambassadors to help them to maintain their brands and create new ads.

‍

Rajan: What part is human? now after using the tool, would you say that earlier it was X percentage and it is reduced by something now like you know, if somebody were to build an ad before a cool, they would spend eight days and like you know, going through it and now much lesser number of people and then they spend less amount of time.

‍

Jiajun: Definitely. For example before if you want to hire a person to record a video for you to make intros or ads and so on, it takes lots of time. It takes like a week or a few weeks and so on. but now if you use avatars to help you introduce products and services, it only takes minutes, right? So it's very fast. As well as if you want to record an ads for peoples from all different countries and or even with celebrities, their schedule is very tough. It's very hard. And now you can use a cool technology to do this very simple or if you want to do it for all the different languages people speak. It's just very very hard and very very difficult process as well at a very high cost. So we help them. Besides that we also enable some ad experience that's not possible before. So before all the ads is one video and it being viewed by millions of people. But it could enable the experience of create millions of videos for millions of people and each people watch their own version of videos. So these ad experience is not possible before and now it becomes possible because we enable it.

‍

Rajan: What's the problem in video AI on the technical side that people said could not be solved before, but you figured out and you're doing it now.

‍

Jiajun: So in video AI the problem is still very difficult. It's still very difficult to create high quality videos. especially if you want the video to be commercial quality and even commercial quality for big brands or movies and so on. It's still very hard. So moo, it's just AI video generation is not there yet to create these high quality videos. So what we did is we combined a lot of different approaches and the different rendering stuff and different components and assembled a system that is able to create high quality results that can satisfy the needs for their ad creation. And I think many people are very amazed by solar and some of other video generation tools. But when really talking to the clients, no matter it's Bollywood or talking to the clients or it's like largest brands, these videos are not to their production quality yet and still long way to go. So what it quoted is we take the approach that we work on the technologies that mature enough to be able to be productized as well as we use a system approach rather than a ah, single technology approach to assemble a high quality system that can serve users needs.

‍

Rajan: How do you know if the ads are working? Are there any specific metrics that you look at?

‍

Jiajun: So we track several metrics and we also work together with Harvard Business Group to track much more metrics to see whether ads are working or not. Some simple metrics will be how many videos are being created, how many people are creating videos or view the videos and how many shares they have and how many comments or if it's a recreation of videos. And we measure how many recreations of the video there is in social media and so on. So there are a lot of metrics, usually data metrics to view how popular a content is in social media or in different places. We use these popularity metrics to measure how successful the ad is. And meanwhile we also work with some of their brands so on trying to help them to measure the conversion as well such as website registers and the purchase. But these metrics are much harder to collect.

‍

Rajan: Let's do a quick round of short question. Just give me the first answer that comes to your mind.

‍

Jiajun: Okay.

‍

Rajan: Coding language that you can't live without. Python books that shaped your approach to AI?

‍

Jiajun: I have no idea. I don't read that much books now.

‍

Rajan: Biggest misconception about AI in ad.

‍

Jiajun: AI is mature enough to do all the ads work.

‍

Rajan: Your go to productivity hack.

‍

Jiajun: Start with simple work first and when you get started you can do more and more difficult works.

‍

Rajan: Your favorite ide, I just use Veeam.

‍

Jiajun: Oh, okay.

‍

Rajan: Wow. Okay. One word to describe the future of AI.

‍

Jiajun: Everything is possible.

‍

Rajan: So let's go back to again the longer set of questions. Let's say it is 2030 and AI is making 100% of all the video to add. What does the day look like for a marketer now?

‍

Jiajun: So I think a lot of the work will be done automatically by AI. So if you look at the current marketing space, especially In United States you will find that a lot of these works are done by agencies. So there are all kinds of agencies from the largest ones to small ones and the agencies take care huge amount of works. And a lot of these works are very scattered and very small. But it needs to be done. So that's how the current marketing works. But I believe that pretty soon a lot of these agencies will be replaced by AI or some of lots of their works will be replaced. So there will be more focus of the creative works and lots of the execution and the creations and edits will ah, be automated. And when that happens there are two possibilities and I'm more likely to choose that ads will be much more personalized. So which means there will be lots of different ads happening and a lot of them are created by AI. So now I think people are so bored about ads and nobody likes ads. But when ads becomes more interesting, I think people will like ads much more.

‍

Rajan: In your product you put faces in videos, right? It is possible that it can get misused. So what are some checks and balances that you have put in place to ensure that you know the technology is not used in a harmful way?

‍

Jiajun: So there are several things that we do to ensure that the technologies are properly used. First of all we review the accounts regularly and if we find some content that's not allowed by our platform, we will block the account. And the second, we recently launched a new feature called Javas Moderator and it's an AI moderator that can moderate content and we are integrating the AI moderator into our system. And the third thing is that we are working on some system to help to ensure the consent of the proper usage of the human videos and images, especially with faces.

‍

Rajan: So if your company was a movie character, who would it be and why?

‍

Jiajun: I like Superman the most so, and why Superman? Because I think we are really building something that's good for the human race and also protecting everyone and enable everyone to have the ability to have the super power which means create a digital version of yourself or handle, you can allow some digital copies of yourself to handle different works and so on. And also I think the mission that it carries and then the mission the company carries are similar.

‍

Rajan: What's the hardest tech problem that you had to solve for scaling your company?

‍

Jiajun: The hardest tech problem? I think in the generative AI space the hardest problem is still about the result quality. So how can you create results with the best quality? It's still the problem that everyone is working on and everyone is trying to.

‍

Rajan: Figure out what's something that you've seen, as a surprising use of your video technology or any video technology.

‍

Jiajun: I think just one week ago, one of the users reached out to us and says that they used us to create a business which is create personalized cartoon books. So for every kid, they can take a photo of the kids and put the kids into cartoon books and let the kids to experience a new adventure and they will print out the cartoon books and ship it to your home. So, I just created a cartoon book for my song and then still waiting for it to arrive, what is it called?

‍

Rajan: I'll link it in the show notes.

‍

Jiajun: So I'll send that, later. But, it's definitely a very interesting one. And I shared on the LinkedIn. I don't exactly remember what's the name though. Okay.

‍

Rajan: All right, let's say you're having dinner with three people from history to talk about AI and creativity. Who are they and what would you ask each one of them?

‍

Jiajun: Three people, I think Einstein definitely will be. So I think Einstein studies a lot of the origins of the, universe and so on. Right. So it will be very interesting to discuss the origins of the intelligence and this kind of stuff. And I think later he was also very interested in how the world is created and maybe how the human is created. Right. So it's a very, very interesting stuff to discuss. The second one, I think it will be Turing. So Turing create, ah, made lots of contributions to the AI as well as computer, science. So, I think he will be surprised if he realizes that the AI is progressing so fast, that something he thinks was very difficult, impossible, and hit time becomes possible just, not too far later. And, what will I ask him? So, well, the touring test is already like almost hacked. Right. So almost gone. So what's the next one? So do you think the intelligence really comes or you want to increase your bar? So, that's about Turing. And the third one will be maybe Da Vinci. So he studied a lot of different things about science and so on. Right. And he also studied the human body a lot. So I think it will be very interesting to discuss with him about, creativity and about his thoughts about AI. And he's a very, visionary person and how he would think the future will look like with AI.

‍

Rajan: So how do you convince big companies to work with startups? What did you do for big companies to work with you?

‍

Jiajun: How to convince big Companies to work startups? I think that's a very good question. So there are several ways that we do to work and get connected with big companies. First is there in frequently lane we work with middleman, right? So the middleman there could be a company in the middle or these kind of things that we work together through the middleman and then together with the big companies. So that's a very good approach which means we have channels to these big companies. The second one is we do have many connections that help us to get connected with big companies and work together. And I think the third one is really make m sure your technology and products are very good. And as long as it's very good, when big companies need something they will look for you. And actually most of our clients are embalmed so which means they look for us.

‍

Rajan: What are some of the biggest mistakes that you made at Akul and what did you learn from those?

‍

Jiajun: So at the early days we did some explorations. At that time I think we tried to immediately go to very large deals and it turns out that it's not a good start for startups for immediate large deals because they are slow, they are competitive and there are lots of issues that startup is very hard to handle. So that's why we make PV and we go to some area that is more easier to get started and something that's helping us to helping us to grow in the process. meanwhile we also grow our customer base and we take larger and larger clients.

‍

Rajan: Tell me a little bit about your tech stack. I'm curious about like what kind of tech stack that you have built for building a video AI startup.

‍

Jiajun: We definitely use a lot of the cloud solutions. We use AWS and Google Cloud and a lot of our basic infrastructure is based on these cloud providers. And on top of that we build our own tech stack which is mainly based on Python and React JS and some Node js. So these are the main tech stacks that we build on and by leveraging their cloud resources we are able to scale much faster for our infrastructure as well as we do have several person focused on optimizing our infrastructure and optimizing all the different stuff in terms of the AI models that we train. And so I think most of them are based on Pytorch and also we write lots of the code ourselves to implement many different system related tasks that's needed.

‍

Rajan: How do you explain your company A cool to a 10 year old just using things that are available in let's say a kitchen.

‍

Jiajun: So we help you to make clone of yourself in a video and animate the videos in different ways and then use the video to create ads and so on. So that's we have create human videos, mailing for ads.

‍

Rajan: Are there any industries that you feel where AI video tech can have a big impact? Anything that you have tried and did not work earlier and may come back later.

‍

Jiajun: Our customer base is mainly four categories. The largest one is advertisement, that's the majority advertisement and largest one. And the Second one is APIs. We have pretty good API service to support startups as well as enterprise. So they start up and call our APIs and create lots of cool stuff. And the third one is film production companies as well as studios. So we work with Bollywood to create Bollywood movies and celebrities and so on. We also do the fourth category is creators and influencers. So these are the menu, the Zhonghao platform.

‍

Rajan: What questions do you wish more people would ask about AI and video? And they're not asking.

‍

Jiajun: I think the thing is people are really interested in this field but a lot of them are not sure how fast things will come and how the future will look like. And many people are too optimistic or too pessimistic. So when it goes to extreme usually, usually the people in the tech field they are too optimistic and so for the other people they are too pessimistic. So yeah, basically what's a what's a realistic prediction how the AI video will go.

‍

Rajan: Last thing, recommend one book, one tool and one habit for someone who wants to understand where AI is going.

‍

Jiajun: One book I would recommend a book I wrote which is called Enhanced Human. But currently that book is only in Chinese, it's not in English yet. So some book everyone can read I like the most is A Brief History of Time which is written by Stephen Hawking. He talks about physics, the origins of the universe and so on. And one tool. So tool I like the most or use the most currently is really chatgpt. So I use chatgpt a lot to help to either answer questions or search for documents or search for staff or so on. So I use ChatGPT a lot so definitely it's a very helpful tool and one habit. So the habit I think that's very important And I'm also learning is need to do a little exercise every day. It doesn't matter. It's half of an hour or something, but a little bit exercise will help you to refresh yourself and work better.

‍

Rajan: Awesome. Jeff, thank you so much for joining, Pivotal Clarity. All the best with your startup and your, future projects.

‍

Jiajun: Thank you for imagining.

‍

Rajan: That's it for this episode of Pivotal Clarity. This is an opaque podcast. OPEGA is an accelerator for global Indian founders building AI software companies. We're exploring the fast changing world of AI together with our listeners. If you like this podcast, you can find more on our website and other popular podcast apps. Subscribe if you want to keep up.

‍