Revolutionizing Personalization: Personal.AI's CEO Suman Kanuganti

0:00The wait is over. Dive into Audible's most anticipated collection, The Best of 2025. featuring top audiobooks, podcasts, and originals across all genres. Our editors have carefully curated this year's must-listens from brilliant hidden gems to the buzziest new releases. Every title in this collection has earned its spot. This is your go-to for the absolute best in 2025 audio entertainment. Whether you love thrillers, romance, or nonfiction, your next favorite listen awaits. Discover why there's more to imagine when you listen at audible.com slash best of the year.

1:03I'm an engineer for almost like first half of my life. Robot experience. I also have a business background. But generally, I pick the passion of creating products for people. I always get excited when you create experiences for people to touch and feel, kind of elevate the day-to-day activities of that movie. That's kind of motivates me and excites me. I created a company called Aira before personal AI. Aira was a technology that uses both AI as well as augmented reality, but with human for people who are visually challenged. Yeah. So today we basically unlock, you know, all kinds of experiences for blind community and, you know, pretty much the entire United States and all the retail stores kind of, you know, provide that as an accommodation for people.

1:50Yeah. I started Personal AI three years ago with a core premise of, you know, being able to create a model to represent one individual and i'll tell you more about it but uh since you asked me about my background yeah that's a little bit about my background very very cool so what kind of got you interested in well actually perhaps give us a little rundown on what personal.ai does for those that don't know um and uh yeah explain what you're what you're shooting to do there yeah so personal.ai is a model it's a language model that is grounded in one's facts opinions and perspectives of an individual.

2:27So it is designed to reflect and speak on behalf of an individual, right? The most interesting thing about personal.ai is you probably, or the audience probably, have long discussions about large language models. We actually started developing the opposite of a large language model. We call them small language models or personal language models. They're actually tiny. They are 120 million parameters. They are grounded in one's memory. These models have long-term memory. They also have attribution back into the data. But obviously, it doesn't know everything that the world knows, like a large language model, but it knows you, and it's an ultra-personal experience to the point where it speaks like you, within your own facts as well, not just style and voice, but also your facts and opinions.

3:23um that's so interesting i think especially today in a day and age where i've actually heard a lot of pushback from people using tools like chad gpt or other large language models saying like the responses it gives me don't actually reflect my opinions it has biases perhaps that came from the company itself perhaps that came from people training it and you know a lot of people kind of complain about that um for all sorts of different reasons right there's political there's religious there's all sorts of ideas that it's really hard to define what's correct what's right or wrong from theological or ideological perspectives.

3:56And so I see a lot of appeal to having something like personal AI. How do you go about training one of these models? If someone's interested in getting a model of themselves, how does that work? Yeah, yeah. No, it's pretty fascinating for me as well because obviously we started this company three years ago and we started working on our long-term memory, the models, the different and transformer architecture that we have now, and then ChatGPT happened. But if you put that in the perspective, open AI, large language models have been in development for like five years or so, and then it suddenly started hitting consumer market after eight years or nine years or so.

4:38So I always tell the people around me, it's like, hey, personal AI started three years ago. That's when we started our models, and we need to make it available for people to train it super easy. Like, you know, what are the systems that you have where without needing a developer, you can have a user trained model, right? I think. So yeah, so we basically developed applications on top of our models where people can ingest any source of data that you would wish. So that data could be mostly authored content, right? The authored content would be your articles, your drive, your existing autobiography.

5:20You can choose to take some of the historical conversations that you have had in other chat applications as well, and then you can use that as well. So it depends on what kind of authored content that you have that surrounds you, that represents you, but you can think of it as if you do have an existing journal or if you do have an existing autobiography or do have existing composition, you can use that to create a model of yourself. And we have a mechanism where you can use the applications for ongoing communications so that the training is not once and done, the training is like ongoing. Because it's a small model, it updates instantaneously.

5:54Very cool. What does that look like from a practical perspective, right? Like let's say I would like to train a model for myself. Am I just going to like a folder and uploading like as much content about myself as I can and the models taking in that type of content? Or like what does that look like for a user? Yeah, so for a user, you essentially have like a chat-like application. So if you could imagine ChatGVD, right? We have our own native applications and you can essentially tell your AI anything that you would want it to remember or to be trained on, right? So it could be text data and that text data could be a simple statement or a large paragraph or a huge article or a huge document.

6:37You can choose to bring in like existing documents and upload them as well if you would wish. We are trying to make the process simple by getting the data from other locations. We have Twitter, for example, like, you know, people who has been collecting or authoring the content on Twitter, you know, has large amounts of content. They can simply connect Twitter and then boom, everything, you know, will be synced or added to your memory. So as long as that content that you think will represent you in the first person and author, if that is your intention, yes, you can push all the data by simply a messaging interface.

7:15Very cool. Do you or are you thinking about the integration of video? Obviously, text is a good way to integrate someone's personal audio. But let's say someone has a YouTube channel or like myself with a podcast. All of my thoughts and opinions that I share on here, is there a way to pull some of that input? Yeah, so the input is as simple as if you do have YouTube videos, you can send the YouTube link to your AI platform as well. so it will extract and then set if you do have like medium articles you can do that too if you connect google and then have gdoc that can be pushed as well uh it doesn't have audio transcriptions uh built in but people generally use like third-party transcriptions and then push the data in okay very cool so who's the main target audience that's currently using this and where do you kind of see this going in the future like is this authors looking to write books is that or is it just people looking to share their opinions?

8:18Like who's using personal.ai right now? Yeah, so we are fairly new to the market, right? And that's one of the reasons why, you know, we are actually having this conversation as well. When I say fairly new to the market, we've been experimenting with people for almost like one year and then shaping up our products and APIs and needs and everything else. So since we went GA, which happened in April, we are seeing a lot more traction on people who are interested in their personal brands and elevating their personal brands, right? So these are solo-preneurs mostly, small business owners mostly. And the most interesting thing is these people have something of them, which is personal data because their personal data is their personal brand and it is their currency, right?

9:04And we are very much focused on individuals. So it's more like a B2B2C kind of a model where people who would want to serve a small group of community, and that community could be 10, 50, 100, or even 1 ,000, they are creating their personal AIs. They are integrating into their existing communities or colleagues or clients, if you wish. And the beautiful part is the native applications also supports native messaging, meaning you can invite friends or communities and create AI lounges and invite people for prompt party, if you will, you know, to talk to your own AI and geek out. So those are all the experiences that we have unlocked.

9:48We just literally, two days ago, released our mobile application as well. Oh, very exciting. And is that called personal.ai? Yes, it is personal.ai. To search in the App Store, since we are still working on optimization of search, it's personal.ai mobile. And then that's the app. or you can simply go to personal.ai slash download and it's a brand new application. Very exciting. Very cool. We haven't even lost yet. We haven't even told the world yet. We are just serving our own people right now who are in the system. Okay, very cool. I'll have to grab that and check it out after. So where do you see this going, let's say, you know, three, four, five years from now?

10:29Where do you see AI or personal.ai kind of growing to and who is it serving? Yeah, so I think one of the core principles that we haven't touched on about personal.ai is every person gets their own memory, meaning the memory data is not shared between two people. And every person also gets their own model. In other words, the model is not combining the data or aggregating the data, which is very popular in the large language model scenario, but everybody gets their own unique model. And this data, which is a memory, as well as the model belongs to the person, meaning they actually own the data.

11:06That means we are not allowed to see it. We are not allowed to sell it, no ads, but they can use their own model if they would want to and then monetize it as well, right? So today, the most popular use case that we are unlocking given how good these models are to represent you is the daily communication, right? So if there is incoming message that is coming from your friend or client or community member or a customer, your AI will draft that response for you and it is since grounded in your own facts and it is learning almost instantaneously every day, it does 70 % of the job for you. And you can choose to either review it and approve to send that communication back, either using native application or, you know, like integration such as iMessage or other places.

11:53Or you can set yourself into autopilot. So if you set yourself in autopilot, that means technically it's Jaden's AI version that people can talk to, right? But it's not just like dependent on, you know, super high celebrities, but it's now available for anybody to be able to create an AI version for themselves. Self-service, a user train. In the future, because of the principles that are attached to this personal AI, which is we want personal AI to be the starting and the ending point of everything. And it is private to them, meaning only individual people will get to decide what goes in it. They will have full control of it.

12:31So we see a future where, you know, future personalization like services, for example, when you open the app, you know, typically you send GPS coordinates or you send information to a service for them to provide you some suggestions. let's say an Uber or an OpenTable or Airbnb, right? We see a world where most of the services can integrate into your personal AI to give you the ultra-personal experience without compromising the data integrity of the individual, right? Yeah, yeah.

13:08So, and I think the starting and the ending point of everything that you will have on the internet presence will be your personal AI model because it belongs to you. That's super cool. And something that you said made me think you said, you know, people will be able to monetize this. Just out of curiosity, have you thought about like the possibility of, right, someone trains or creates an AI model of themselves and then like having them being able to have like an API so other people can like ping off of there? Is that something that is future or is this more close? Now, that's part of the architecture.

13:39So think about every person, right? Every human being at center. And then surrounding them is their long-term memory. And that memory is time-borne, meaning as you go by your line, your memory is growing and you can capture it at every period of time. So it's all like time-borne memory, okay? Now, the model, which is personal language model, is an ensemble of many different models that will be trained, which is grounded on your long-term memory. and each of the models has an API. So think about that API is API to your model, which is API to you, right? Yeah. Now that API is what is used for our personal AI native applications, but a good number of people today are using that API and taking their personal AI model in the places where they already are.

14:25Either it be communities or websites or Discord or Telegram, they're already doing those integrations. So that way you have your AI in the mix in these communication settings. now you can think about using the API for your agents as well so agents are the ones who are taking tasks or doing tasks or you know like actions for example drafting an email for you right so that's creating a tweet for you that's another agent being an email for you another agent but at the core it's still you it's your memory it's your model that's the presentation of you that's so interesting it just it made me think of how cool it would be if um like let's say you take like a company right so you got like 20 people in the company all of them go and make like a personal ai for whatever they're like for themselves and then you you try to like test yourself against the ai version and you like get the you like get all the people to start like an ai company essentially where there's all the different employees trying what happens in our company every person in our company so by the way we we i mean of course we building personal AI.

15:31So every team member has their own AI. Yeah. We all have our own lounges, right? I have a Suman auto-pilot lounge. So customers and people actually who come in, that's where people go into prompt with my AI. So technically, whatever is that you are asking, Jaden, you can interview my AI version. I should have just gotten that on the podcast. I could have saved you some time. I sent you the link. So, s.personal.ai, so you can technically simply go in there, click chat with me, that will come to your application. This is a human in the loop experience. So, in other words, Jaden, we're talking to what now, 15 minutes or so?

16:14Yeah. There is an element of trust that we develop, right? It's like, okay, I know Jaden. When he comes over, I will start having in the co-pilot experience, meaning if you ask me like someone like how's it going my ai will draft a response for me and they'll choose to send it but when i put you in autopilot you can just simply go back and forth with my ai in a conversational mode um so yeah so you know other uh small businesses right now are essentially building every individual's ai and and the beautiful part in here is you know i mean we have seen the ai versions of people out there right there is elon musk ai there is karen's ai there is like influencer people's ai yeah yeah primarily dependent on large language model it takes like extreme amount of effort to be able to ground it it still has right our goal is to make this almost accessible to every person right not just influencers because influencer models are possible today because their data is already public and it's already indexed by large language model.

17:24Right. Nobody knows, Suman and Jaden, you must be popular, but still demo not. Yeah, I mean, but you have your personal mode. Like you have your personal plan. Like AI chats is your brand, right? So how do you invent that brand? Like, so that's where we want to make it accessible. So now you can have your own model of yourself. I love that. That's so cool. Okay, so it's giving me so many questions. This is actually such a cool AI project. This is one of the cooler AI projects I've covered recently. So this is my question for you. Number one is, what does that look like? As far as like the resources, the bandwidth and everything required for training, like one person's AI model, right?

18:04I know right now there's like this huge shortage of like A100 chips from NVIDIA. Everyone's like, you know, fighting over trying to be able to train AI models and stuff. what's the resources you know realistically let's say you had like um everyone on facebook wanted their own personal ai so you got to get a billion people to train their own like what would that look like for resources wise yeah and that's where the best part is almost like the secret sauce and i cannot wait to cannot wait to be out there and give this to millions of people because you know why these 120 million parameters are super cost effective because they are small they scale horizontally we don't need to fight big tech for all the gpu compute in a way i would also make the statements such as like they're environmental friendly because you know why they actually run on your mobile devices right so as of today these are 120 million parameters uh like in less than one gig in size and we have identified few chips to essentially move these models onto the edge itself or your computer and your phone.

19:05That does not have dependency on the large language model. That means you don't need all the investments that otherwise you'll have. So, yeah, because, I mean, just think about it. What we are trying to achieve is our use case is not creating a model of world's information or the internet. Our goal is to create a model of one individual. It's a 120 million parameters. yeah i love that especially because i feel like i've heard a lot of people of course complain that chat gpt just sucked up everyone's data and now it's you know yeah i didn't really so important like yeah so so here's the fun part right you know i got i got into creating this project three years ago because i had this like extreme desire of being able to communicate how conversation with a person who taught me a lot and he passed away because of pancreatic cancer So there was this constant mantra of like, what would Larry do?

20:03Or, you know, I wish I could talk to Larry's AI, right? Because, you know, we as human beings, we establish trust with a group of people that are surrounding us. So even if you are seeking advice or getting an opinion, it's much more about researching and being able to land to your own conclusion versus being able to talk to one or two people that you trust and getting their opinions. so the idea of like accessing people with you know each other to an extent where we are not just bounded physically by the time and the capacity that we have is mind-blowing right yeah um that is man i lost my train of thought so that the reason like this this is all happening uh is because fundamentally and principally we said this needs a memory large language model cannot solve it.

20:53And my co-founder, CTO, MIT graduate, super, like creating like five different AI startups, Sharon Zong, she had the foresight saying that, so one, if you really want attribution, if you really want it to be private, if you really want it to be grounded in one person, that doesn't hallucinate. This was 2020 because nobody knows large language models there. She said, large language models is not the way to go. We had to start from scratch. There are a lot of existing stuff we reused, like open source model, transform architecture, et cetera, et cetera. But it is a contrary approach to the modeling, if you will.

21:31Yeah, for sure. I mean, I can even just think of a cool use case where let's say you have to make a really important business decision or family decision or life decision, and it's very time sensitive. You call up your, maybe I'm calling up my dad to ask him for advice. He's busy. And then And I'm like, oh, shoot, he's like off in another country. He's not going to be on, but I got to make this decision right now. So I just go over to his AI chat. This thing might not be like dead perfect, but I could get a pretty good idea of what his advice would be in a time-sensitive way when a person isn't actually physically available.

22:01Super interesting. Really, really cool idea. What are you looking at as far as monetization, pricing? How does that play in? What does that look like for this? Yeah. So we are making it available for anybody to create their memory and train their AIs. for free at this time. And the intention is everybody gets like two personas, like personal and professional. Personal is nothing but a part of your life, right? For example, you know, you and my relationship, let's say, is professional, but you have, you know, your family relationship is like personal, right? You can choose which category or which persona that you would want other people to access when they are talking to you in autopilot or copilot, right?

22:50So normally people who would upgrade to a$40 per month subscription plan, they have things or access such as greater integrations, greater API access. If you do have a lot more following, like not those 100 people, 200 people, where there's a lot more demand to the AI, then it goes into a little bit more custom pricing as well. So currently it's accessible to everybody. for free to begin. And then there is a$40 per month pricing for greater integrations, fast faxes, and taking to other places. And then as the demand for the model itself grows, then it goes into, you know, kind of custom solutions.

23:32Very cool. I love it. That's super interesting. So I know you've been working on this for the last three years or so um first off what kind of what were you what kind of made you decide you wanted to take this step on this project i know you mentioned larry was that kind of what played a role into into creating this yeah totally i mean that was the denises uh to be very specific uh the project i kicked off like in september september 2019 is when i started working into just understanding what I'm dealing with. Oh, Jesus. I mean, like, I didn't know where to start. It was, like, very vague. And back then, it isn't a space where a lot of information was available even.

24:18Yeah. You know, I mean, I knew I needed AI as a technologist and an engineer, but I myself didn't have, like, hands-on AI experience itself. So first six months is all about, like, researching, you know, what goes into it, like, understanding memory, understanding Transformer Architecture, just reading a whole bunch of stuff and then trying to find my team when I decided, okay, this is what we're going to do, which is try to figure out a model that we can build for people to talk to, right? And that had been representing the person. And then that's when I found Sharon, my CTO, which is my co-founding member now.

24:55And then for six months is, again, like experimenting, researching, talking to people, what does it mean to them, ideating. We took almost two years to put together the core foundation systems, the models, putting together the API. And then we took another year to create applications on top of it and test with people. So now we are ready to go to the model. That's very exciting. How did you meet your CTO? I know a lot of people are interested in new startups or taking their companies to the next level with AI are kind of interested. What was your process on that? I mean, it was very deliberate because previously I was already running a startup.

25:37So I had some ideas on, you know, the kind of person that I would want. Yeah. Specifically for a company like Personal AI. Because one, it's original thinking. So it requires a little bit of like original understanding of the technology and working towards a solution that may not be widely popular or available. So we need like novel thinking. So I went through a search process, believe it or not. I had some friends, recruiting firms who accepted my request to find a co-founding partner and I spoke to 120 people. Holy smokes. Wow. and I showed her the spreadsheet after I got her. That is the process I took, but I took my time.

26:29I took my time to find the right person. My other co-founder is Christy Kaiser, so I worked with her at my previous company. She's my design partner because it's also required a little bit of what does it mean to be talking to the AI. Now it's pretty obvious how they chat with the AI, right? but we created these chat interfaces to basically train and you know chat with somebody's ai and then chat gpt happened uh and then it was it was great because there's tons of awareness and we were heading in the right direction then you know we had to uh we had to double down in the conversation in the chat yeah that's super yeah that's super cool um what did what did that kind of look like i mean you've been working on this thing for a number of years and all of a sudden chat gpt launches you know kind of the the end of last year um was that like was that did that feel good was that like oh my gosh so much validation we're going in the right direction like what was i guess your idea is this kind of went from being like ai was like kind of a trendy thing in like tech and stuff and then to all of a sudden like it is like everything yeah no i mean it happened so fast I mean, I had to like, you know, sit and kind of on the hindsight, think through every week what was happening in the space.

27:48But I can tell you that things were happening so fast. And there's a lot of momentum, a lot of awareness, a lot of credibility as well, a lot of appreciation for the amount of work that we were doing for years. Right. Yeah. because there's one thing about like building the product building the application there's another thing about like um trying to like tackle something unknown and then put together a sequence of execution steps for you to be able to figure it out you know because you make mistakes as well and you have to pivot and you have to change things uh so i guess what could i say um chat gpt and that blowing up was good because there is a general acceptance about the ai like even one year ago at the same time when we were pitching like personal ai to investors or anybody else is like yeah it's like what are you talking about uh but now it's like oh okay that makes sense and what are the applications right like you know what is the mode yeah but at the same time there are tons and tons of AI companies too, you know.

28:57Now, every big tech, every company wants to create personal AI. And our definition of personal AI is a little bit unique. Like, you know, we want it to be true to what it should be, which is representing a person and that is owned by them. It's not just a merely an assistant that can do like one or two tasks that you can talk to a AI bot for a specific reason. That's all good. There are use cases for it. But for us, personal AI as an assistant is underselling. I think there is a lot more beautiful thing about this core idea of having this as an asset to you that belongs to you, that lands over a period of time, that is not compromising the principles of data ownership, that is not aggregating with everything, that does not belong to the company, that belongs to you.

29:49So there is a lot goes into making a true personal AI. And so the downside of this whole bubble of AI is the market is confused, right? Because there is a lot more money that is getting poured in. But sometimes when you have money and you don't know what to do, you would also confuse the market. So you create chaos. And I do feel there are players in the industry that is creating chaos. we have to navigate through the chaos right because we still have to stick to what we think is the truthful way of creating a trustworthy AI that belongs and serves people right so yeah so anyway so I think there is good but there is also the downsides of whatever is happening because now you got to figure out what is what and who is who and we are you know instead of confusing the market we got to work towards making that clear as well one of the reasons i want to hop on the podcast and talk to people yeah i love it um okay so what one question quick is you you talked about you know there's some companies that aren't doing it the correct way right there's a bunch of money coming into the industry and there's there's players like you said that are creating chaos who are those players that you believe that are fundamentally or even if it's not a specific company what the concepts that you believe are fundamentally creating the quote-unquote chaos or kind of pushing it in the wrong direction uh okay we are getting a little bit political um i would name the companies but i will tell you what i think is not right yeah what i think is not right is misrepresenting an identity of an individual whether you are seeking permission or whether you're not seeking permission, meaning you have to have permission to represent somebody's AI or somebody's identity.

31:51Because what makes human human is our identity, right? Right. Now, the thing is, it makes like good entertainment, good business, good fun. But at the expense of that identity, are they okay with it? if they're okay with it that's great but I think there needs to be some empathy and some respect to what we are doing in the industry and I think if we are not doing it right then it's unethical now investors are you know business people will say who cares I don't care I'm like whoever I want to make some money let's get people to engage with it who cares about privacy who cares about identity do it you know that's that's their business we are not going to do that business but i think it's okay yeah yeah yeah that makes sense uh and i like that concept because it definitely and i think to be honest i think a majority of the population now not necessarily the owners of a lot of the ai companies and not necessarily as you mentioned the investors but i think um there's probably a majority of the population that would agree with your sentiment um which is people when they're creating this content if someone like open ai sucks and all the data from the whole world and like i can go to open ai and say you know in the voice of stephan king write me a write me a book outline okay now write the book now write the chapters and all of a sudden it's like well you know stephan king never actually was able to monetize the usage of that is that ethical right then there's that whole question i think a lot of people would would agree with you that it would would like to to monitor or to um you know compensate those people i'm curious though from from your perspective um or from your guys's story what did that look like when you were starting uh personal.ai is this something that you've bootstrapped or self-funded did you get investors from the beginning what did that look like for you as far as funding this company oh yeah i mean we do have investors but our investors are aligned with our principles yeah so first year was self funded so basically me and char uh by second year we needed uh you know team and to build jenks so that is 2021.