TBC’s Alex Ksendzovsky: The Neurosurgeon Using Brain Cells To Build Better AI

0:00Alex Ksendzovsky:What if I told you that the tech behind the best, cheapest, and fastest image and video models might not come from fancy AI alone, but from the same kinds of cells living inside one of the oldest technologies of all, your brain? It may sound like science fiction, but at the Biological Computing Company, a startup that came out of stealth earlier this year, it's becoming reality. We're not building brain-inspired AI, we're building biologically-derived algorithms, right? And so this is a bit of a stepping stone towards real-time biological computing. but it builds the foundation to get there. This episode is brought to you by Rippling AI, the only AI built to give you full visibility into your startup and the ability to take action across every department.

0:41Alex Ksendzovsky:On this week's episode, we discuss how co-founder and CEO Alex Zanzowski went from neurosurgeon to tech entrepreneur, how human insight and living brain cells can help power a new generation of AI models and why he thinks the data center of the future just might be alive. I'm Alex Conrad, founder and editor of Upstarts Media, And this is the Upstarts Podcast, our weekly show where we talk to emerging startup founders who are punching above their weight and taking on the status quo. Alex, thanks so much for joining the show. Thanks a lot for having me. We've got an Alex Alex pod here in San Francisco today.

1:12Alex Ksendzovsky:What is the biological computing company besides a very sort of intriguing name that I think you guys settled upon after another one? Yeah, yeah. So I think the simplest way to think about it is we are building software from wetware. So what that actually means is we take real biological neurons, so real brain cells, grow them in a dish of electrodes, and we build software tools that emulate how the brain processes information. And so we're creating algorithms that are a lot more efficient and have much better quality. And so you're working with these neurons that are living neurons that can die.

1:45Alex Ksendzovsky:And from them, you are essentially able to create better models that can be used for images and videos, more akin to how we would think of sort of traditional AI software. Exactly, yeah. Yeah, so the product we built itself is algorithms that are more efficient. We're starting with video generation, so the output is much more higher quality, but that can extend to language and other modalities as well. The premise of the company is that the brain processes information in a very special way. It's a lot more efficient in the way it processes it. It's a lot more dynamic and complex. And what we do is we encode information from the outside world into the biology.

2:20We observe how the neurons, the brain cells process that information. We mathematically model that. and build adapters, learning rules that can be incorporated into current AI systems to make them better, faster, cheaper, stronger.

2:34Alex Ksendzovsky:So I know you've only been out of stealth for a couple months. You guys announced a seed round of$25 million. Is this in the market today? Do you have customers? Or just how early are we with TBC? We have a product that's a TBC-enhanced open-source video generation model. We have a partnership which we'll announce in the coming months that's serving this model on a neocloud. And so we're going to ultimately, customers will be able to use it on the marketplace. And so, yeah. And so in the coming months, our product will be deployed and people will be able to use it. So when I first heard about the idea of, you know, neurons helping to train AI, I'm sure my mind, like a lot of people, went to, oh, like, are these neurons in a data center?

3:16Alex Ksendzovsky:Are we going to see racks of like, you know, living GPUs or something like that that are creating a much better output? But that's not quite what's going on here. Can you correct for everyone else the mental image that they might have? So not yet. We're getting there. We're building the foundation to perform what you're suggesting is real-time biological computing. But to get there, it's going to require probably five, ten years, and a lot needs to be figured out. What we realized is in the interim as we're building this foundation, we can really incorporate our findings into current-day AI systems, right?

3:49And so we can make significant strides on transform models, for example. We can build new algorithms that mimic the brain more closely. We're not building brain-inspired AI. We're building biologically-derived algorithms, right? And so this is a bit of a stepping stone towards real-time biological computing, but it builds the foundation to get there.

4:09Alex Ksendzovsky:What is the big difference between inspired and derived? Why is that so important? So there's a lot of interest, it has been, for many years to try to emulate the brain and biology and build algorithms. But everyone stops short at this idea of inspiration. And so, for example, you look at the hippocampus from memory, you look at larger circuits in the brain, ask, okay, well, how does behavior happen from these larger circuits, okay? But there's so many different things that happen. There's so many things that need to be true from a biology perspective to be able to have a behavior occur in the brain.

4:39And so ultimately, other companies have fallen short because they can't actually model, can't actually see what happens in the neurons themselves. So we have this model of neurons in the dish. We can actually interact with them and code information into them, understand how they process that information. And so we're actually deriving these principles rather than just reading the literature like everybody else's and trying to figure out what the important concept is.

5:01Alex Ksendzovsky:Now, you are not originally a Silicon Valley tech founder. You were an academic on the East Coast at the University of Maryland. How did you kind of end up a Silicon Valley entrepreneur from that Baltimore start? I've been obsessed with this idea for over 20 years. I wanted to do this when I was a college student in 2005. I worked in a lab, one of the first labs growing brain cells on electrodes, and immediately thought that we should be using it to predict the stock market. That led me onto this journey. And at the time, I kind of pitched it to my professor, and he said it was a bad idea, and he was right.

5:36Alex Ksendzovsky:What was your goal with that research? Like, were you thinking you were going to become an academic yourself, or were you excited by research? Why had you ended up in that lab in the first place? He came and gave a talk to one of my philosophy classes, and I was just amazed by what he was doing. He was trying to understand how memory works and individual neurons. So I was just amazed by what he was doing, and I just had to gravitate to that lab. Something just brought me there. Within a couple of weeks working there, I started to think about how do you apply real world to these biological cultures rather than just kind of understanding mechanistically how memory, creativity, and other things work.

6:10Like, how do you apply a real application to this? This has been in the back of my mind ever since, but I ended up going to a career in academia through neurosurgery, so practicing neurosurgeon, and also a neuroscientist researcher, where I was kind of filling my toolbox to get to this point. So my research efforts in my lab were growing neurons and electrodes to study epilepsy and disease. Through my clinical practice, I'm the type of neurosurgeon who places electrodes in the brain to try to understand epilepsy. All of these concepts come together and converge in a dish of neurons where we can abstract from the neural signals, just like we do from epilepsy patients.

6:47Keep the cells alive for a long time, just like I had to do in my lab. So my research, my clinical career kind of converged on this 2022 moment where I was like, okay, now the technology exists to do this. And so like you said, we started in Baltimore. My co-founder, John, and I, my best friend, we started another company together. So we had this realization that if you have a really good idea, you don't need to push it through the academic route. You can actually get venture capital. You can actually and get other folks to fund it. And that's when we started the company in 2022. And so it was really born out of this idea that we were obsessed with this concept and it had to exist.

7:24And we were finding the best way to get it out to the world and try to understand it and try to create it. And it just so happens that Silicon Valley, San Francisco is where crazy ideas are born. And so that's how we ended up being entrepreneurs.

7:37Alex Ksendzovsky:You know, I talked to founders who grow up wanting to be an entrepreneur, maybe their parents are small business owners, or maybe they read about startups and they're really excited. Then I also talk to people who become really deep domain experts and to advance their field or to commercialize it, they feel like, oh, I've got to go do this myself as a founder. And they kind of back into entrepreneurship. When you were growing up, were you like, I want to be a doctor? Were you like, oh, startups are cool? What would have been your journey from the lab to the startup? Yeah, I think it's more the latter.

8:11You know, when I was growing up, I wanted to understand the world. I wanted to understand the brain. I wanted to understand what makes us who we are. That led me towards a path of neurosurgery from a clinical perspective of having access to the brain that ultimately led me towards research and neuroscience. You know, as I mentioned in college when I kind of became obsessed with this idea of biological computing, that was a nice route into where we are today, where it's an idea that I feel like I need to solve. John, I need to solve this for the world. And so it was more of an obsession around this concept, and we backed into entrepreneurship that way.

8:46Alex Ksendzovsky:There's a story that I know you've told multiple times where you and your co-founder are in like an Airbnb in Washington trying to get neurons to see a pattern with S &P 500 data. Why were you in this Airbnb? Like, give me some context there. The S &P 500 idea. So this was kind of a nod to, you know, I was a poor college kid. And, you know, the obvious use case at the time of biological compute was to predict the stock market. Like it was like so clear to me. Yeah, like we can, yeah. Obviously, it's not no longer what we're doing right now. But when we decided. Right now. For sure. Well. Maybe someday you'll circle back.

9:24Yeah. But what we landed on when we decided to start the company was we needed a data set. Right. And like, why not? Right. Why not try to, you know, John has a finance background before he went to medical school and did neuroscience research. He was actually on Wall Street. And so we're like, well, we need a data set. Why don't we start to encode S &P 500 values into the dish? You know, when we started the company, we raised a small pre-seed round. So we were, you know, we were looking for lab space. You know, John was an Airbnb. I can't remember why in D.C. at the time. I was in Baltimore. And so we just kind of got a bunch of computers together, started to look at S &P 500 values, trying to understand which data to encode.

9:58ultimately were able to raise enough funds to open up a lab in Baltimore, independent of my other efforts. And that was the first data that we encoded. And so, kind of a bit of a nod back to history. We got some pretty interesting results, honestly. I'm not going to say that we predicted the stock market, but it was pretty cool. We pivoted onto other data sets that made more sense. But yeah, that was the start.

10:19Alex Ksendzovsky:When I heard, okay, they were able to get these neurons to spot a pattern in the market data. My next kind of question, which might be very basic, but, you know, I'm probably not the only one who had it, is what exactly would that pattern look like? Why would that be valuable to anyone besides just like, oh, here's a spike of activity. These neurons fired, you know, at a certain time. Well, it's a prediction. It's a prediction of what happens next. And so think about it, think about intuition. What is intuition in the human brain, right? And so as we're sitting here today, you know, your brain is constantly predicting what the next thing that's going to happen in front of you is.

10:53Similarly to kind of old school stockbrokers when they're staring at charts, you know, the thing that makes a lot of them so good is the intuition and the gut feeling of what to invest in next and when to sell, buy, et cetera, right? And so what actual intuition is, is your brain and the way neurons are firing in relationship to one another, predicting what the future is going to look like. And so that's what's happening in the dish of neurons. So there are patterns that are emerging that we can decode into real world actions. You know, when we were in stealth for the first couple of years, we were really building a language.

11:24And kind of what you're asking is, how do you know what the signals mean? How do you know what the patterns mean? How do you know how to send a pattern to the dish?

11:30Alex Ksendzovsky:Or how did you know that the signals would be valuable? That there would be business insights or model improvements that would actually create a product from that? That was our intuition in a funny way, right? So our intuition is that there's value in this, right? And for the first three years, we had to build a language to encode information and decode, extract that value. And now the whole premise of the company is how do you apply it towards these real-world business use cases? Throughout the first three years, we were getting results that were incredible, showing that we were right, our hypothesis that in an extremely energy-efficient way, you can pull out patterns from this dish of neurons and apply them towards real computational use cases, whether it's prediction or it's classification or now it's building software tools to improve video generation.

12:14There are the special properties of brain cells that we can now extract, model, and apply. And so, yeah, so the whole premise now is now that we have a language to encode information and decode the neural response, now we're applying it and showing significant improvements in video generation.

12:30Alex Ksendzovsky:We're going to talk about that in a sec, but I think first with painting the picture of these neurons, you had done this in your academic labs, you probably knew how, but not a lot of startups are playing with live neurons. Where are these neurons coming from? How do you keep them alive so that you can, you know, in a viable way, actually train things off of them? Like, give us the high level of what that setup looks like. Yeah, so neurons come from either animal brains or rat brains, or we can create them from stem cells that come from human skin cells. And so we have both models. There's advantages and disadvantages to both.

13:02We grow them or maintain them on what's called a multi-electrode array. And so I know you have yours here. Yeah. You're just in the lab.

13:08Alex Ksendzovsky:So we went to the lab this morning. We got a tour, and I got this souvenir. And what are we looking at here? So this is the multi-electrode array. So this is where we maintain the cells. And right in the middle of this dish is our 4 ,096 electrodes. And I keep referring to electrodes, and I don't want to go too off topic here, but just to tell the audience why it's important. Right now what's happening in your brain as you're listening to this, as you're watching this podcast, is the light waves, the sound waves, are being converted to electrical signals in your brain. That's how your brain understands and creates images, sounds, et cetera.

13:40When you don't have those senses in a dish, you need to communicate with the neurons somehow, and that's through electrical signals. And that's why it's important for them to live on these electrodes. So, painting the picture for you, as you asked earlier, we take brain cells from rats. We create brain cells out of stem cells. We keep them in a perfect environment. And so, they're at body temperature 37 degrees. A certain amount of CO2 needs to be maintained. They live in a pink media. That pink media maintains sugars, amino acids for them to live. They do produce waste, just like any other living cells.

14:13And that has to be removed. So we pipette new media in or new food in and remove the waste every couple of days. And so there's a process to it. I mean, you have to take care of them and maintain them. They live for many months, so six months to a year at this point. But to get here, it required massive optimization. And so just a Quick aside in terms of, you know, academic world versus the startup or industry, our biology team is maintaining the neural cultures. Everything is aimed towards how do you make them live longer? How do you get the perfect culture? How do you get improved cell viabilities?

14:44And so over the years, with an incredible team, we've been able to do that. And so they live for, like I said, many months. We can do many targeted experiments in terms of encoding data, pulling out these, extracting these software adapters for our algorithms. And so we've been able to scale at this point.

14:59Alex Ksendzovsky:Now, you'd already trained to become a neurosurgeon. You were doing academia. These are not easy career choices. Was launching the company hard in its own way or was it relatively straightforward compared to all the work you had done to get to that point? When you think about sort of what were the hardest moments here, where does just launching TBC stack up? It was about as straightforward as this, right? It was not easy and still not easy, right? I mean, we're bringing something completely crazy to the world, right? And we're showing that it works, right? And we're still battling with this. One of the hardest things is convincing people that this is not a research project, that this is production ready.

15:39We have models built off of our biologically derived concepts that are better, right? That have better evals, better benchmarks. And we have to convince the world, right? I think that what we're doing is so unique that there's a lot of interest, but there's also a lot of scrutiny, which I enjoy. I think it needs to be that way. And so to get around that, a big part of our launch was a blog that we put out. And I encourage everybody, go to tbc.co.com because that's our data. And we will continue to update it. We want people to look at it. We want people to understand it. And, you know, that's getting past this kind of continued upstart moment, which is, you know, are these guys a research project versus they're actually building products?

16:22Like I said, in the coming months, we will be deploying our first TPC enhanced video generation model. And so I think that will certainly help. But yeah, that was, I think, the challenge that we've been overcoming for the last couple of years.

16:35Alex Ksendzovsky:These days, you can chat with AI about almost any business problem. Rippling AI is built to actually solve them. That's because Rippling AI is built on your live workforce data. That gives you full visibility into your startup and the ability to take action across every department. Say you're planning your next few hires. Just ask Rippling AI, what would it cost to add two more engineers this quarter? You'll instantly see a breakdown of comp ranges by level and location. So you can make the smartest hiring decision based on where your team has gaps and where the talent is. But it doesn't stop there.

17:07Alex Ksendzovsky:Rippling AI can draft the job wrecks, route them for approval, and kick off the hiring workflow. All you have to do is tap confirm and then get back to building. Don't settle for AI that's all talk. Head to rippling.ai slash upstarts and get AI that turns insights into action. That's R-I-P-P-L-I-N-G dot AI slash upstarts. Sign up for exclusive access today. I assume when you first pitched investors, it wasn't quite this polished, the product you were demonstrating. Was it relatively easy to get at least some weird investors on board right away? Or was there a breakthrough or like a key demonstration that you guys had to nail for, you know, investors to give you money to start hiring people?

17:51So we actually pitched with no data. No data? No, yeah, because we needed separation from an IP perspective from any laboratory and obviously the way we, or any institution, right? And so obviously the way we obtain data is through growing neurons. And so we pitched with no data, but we had a great plan. John and I have the background to do this. Shout out to Refactor Capital, Zoll, Better Call Zoll, invested in us, believed in us, along with Wonder Ventures and some other great investors for the pre-seed round. And yeah, it kind of kicked us off to get the results. Let me restate that.

18:28Alex Ksendzovsky:Okay. It kind of kicked us off to get the data and results to ultimately raise the seed round from primary. Got it. And at the time that Primary invests, you get the$25 million, now you have data to show you can do the demonstration or some version of it of what you did for me today? Well, yeah, yeah. So the Pre-Seed was an investment that allowed us to really build out a lab, build this language to encode information. So we had tons and tons of data coming in. You know, the thing that kind of transitioned us into the investment from Primary was actually a use case, right? So we went from this idea of, okay, we built this language to encode information, but now what do we use it for?

19:05And so we talked a little bit about the classification of MNIST and CIFAR and the data that we also have on the blog. And so there was very clear at that point that we can build computational tools. We can incorporate this into AI algorithms and improve them. And so that was what led into the seed round. And so certainly now we have capital to come to San Francisco, build out the team, deploy this to customers, which will lead itself. into the next fundraising round.

19:32Alex Ksendzovsky:When you were trying to hire that team, you're competing against, you know, the leading AI labs in the world here in San Francisco, a bunch of other cool startups. Were you selling people on like neurons? Like it's alive, like loving the sci-fi part of this or just the practical work that they could be doing? Like, hey, you get to be in this lab or you get to be playing with the models. What was like the pitch that resonated with those, you know, early hires? I think this is the coolest job in Silicon Valley in the country. I mean, I think that our AI researchers are able to really, this is really the bleeding edge, right?