Ismael: 0:00

I would say that there's a universal problem in the space that we continue to see, and that problem is users tend not to want to pay for privacy and security until they get burned by it.

Joeri: 0:15

Hello everyone and welcome to the Web3 CMO Stories podcast. My name is Joeri Billast and I'm your podcast host, and today I'm excited because I'm joined by Ismael. Hey Ismael, how are you doing?

Ismael: 0:29

I'm wonderful. Thanks so much for having me here, Joeri.

Joeri: 0:32

Really happy to have you. Ismael Guys, if you don't know, Ismael full name Ismael Hichon. Rezae Sada, all correct, rezae Sada. Very well done, Sada All correct Reza Sada Very well done.

Joeri: 0:45

He's the co-founder and the CEO of Lagrange, a company focused on decentralized proof generation for roll-ups. With a background in AI and blockchain, he led John Hancock's crypto practice at just 21 and developed decentralized insurance and reinsurance solutions. After transitioning to venture capital, Ismael honed his skills in identifying market opportunities, but his focus is on building. Decentralized technologies led him into yeah, to found Lagrange. So, Ismael, let's dive straight straight in. You have a really interesting journey, yeah, at in from AI and blockchain at John Hancock to founding Lagrange really fascinating. I'm curious what was the pivotal moment that made you shift from corporate and VC to building yeah, your only centralized solution?

Ismael: 1:43

so I've always loved building businesses.

Ismael: 1:46

My first company I founded in university was focused on AI in the context of diagnostics for various psychiatric disorders, and when I moved to work in corporate, specifically in digital assets at a large trad-fi company, my hope was that the impact I could have by implementing an innovative strategy there for digital assets would be much larger than I'd be able to have if I just started my own company, and that was what drew me originally to trying to lead and build a digital asset strategy at a trad-sign company.

Ismael: 2:20

Unfortunately, what I found was that generally, strategies within large incumbent trad-sign companies to embrace emerging technologies face massive regulatory hurdles and these companies won't generally be willing to productize or commercialize the infrastructure, the research that is being pursued internally.

Ismael: 2:41

And so what I wanted to do is be on the cutting edge, and what I wanted to do is build something that I directionally believed in, but I didn't want to do so in an environment where I could possibly be constrained with my launching or commercializing of that technology, and the only way that has full agency over what you do and what you build is to run your own company. And so it really became kind of this foregone conclusion that the only way I'd be able to have, in the end, full agency over what I was doing and the products I would be able to launch and commercialize was through running my own business, and we were very, very lucky early on to have been able to attract some of the best research talent in the space across zero-knowledge proofs and the intersection of zero-knowledge proofs and a bunch of things such as authenticated data structures, ai, proof generation, et cetera, and then to be able to have some great backers who'd support us from day one throughout this journey.

Joeri: 3:37

Wow and yeah, just for my listeners to understand that, like Jeroen is working on decentralized proof generation for ZK rollups, so I have an audience. That is, you know, web3 mixed. Can you explain a bit in, let's say, simple terms?

Ismael: 3:53

what it is about. Lagrange focuses on yeah, so Lagrange. Sorry about that, but Lagrange focuses on proof generation for AI, x, crypto applications, as well as for DeFi and NFT projects, as well as for some GK roles. So we recently launched a product that we call DeepProof, which is the first zero knowledge machine learning library that is able to make zero knowledge proofs practical for AI applications, and so we've announced fantastic partnerships over the last four days with Quill, ai, open, ledger, ungate, and we have a number of other ones coming out over the next, let's say, week, two weeks, and really the use case that we can offer is very simple.

Ismael: 4:35

It is you have a computation of some type, right, you can think of it in the blockchain case, you can think of rollups, you can think of it in the AI case as an inference, and what you want to do is prove that it was done correctly. I'm running it and I want to prove to you I did it correctly. I want to change an asset position on chain, I want to use the result of this inference in an autonomous system, or I want to prove the state transition of some execution environment, but what I need to do is to assert to potentially open or untrusted party or party who doesn't trust me that the result of the computation is correct, and that's what ZK proofs do very well. And Lagrange is a business that specializes in the generation and then the creation of provers that are specialized in the generation of proofs for a broad variety of applications.

Joeri: 5:25

That's a great explanation. But if of explanation, I would say it like that. But if you would need to explain this to your neighbor and ask Ismael, what are you doing? How would you tell it? What would you tell him?

Ismael: 5:39

We let you trust AI.

Joeri: 5:42

That's a good one, because I'm all the time talking about the convergence of AI and your blockchain, and if people ask me what is, then why? It's all about trust, absolutely. I know I read something about your about your infinite proving layer enabling anyone to prove anything at internet scale. Can you maybe talk about some real world use cases and maybe beyond crypto?

Ismael: 6:11

Yeah. So I'd say one of the areas that I'm personally very excited about for machine learning and ZKML is deepfakes. How can you enable a model to be able to classify an image as a deepfake or not a deepfake, and how can you then use the result of that inference to deepfakes, biased media and biased reporting of events? What AI lets you do is both very quickly create and generate potentially large-scale malicious content, as well as to be able to systematically detect and filter large-scale malicious content. The core of doing that safely at scale is verifiability. How do I ensure that if an image is classified as a deepfake, it doesn't show up anywhere? How do I tag it? How do I put it in the metadata? If an image is classified as real, how do I tag it? How do I verify it? How do I ensure that that image is not being spread or dispersed across the world? That's fundamentally what you get at with the question of verifiability in AI and information, and that's what zero-knowledge proofs are the best suited to him.

Joeri: 7:39

But still zero-knowledge proofs, still zero-knowledge proofs. Set key pace. Short, it sounds highly technical for people. It is. So how do you see them bridging the gap between, yeah, Web2, Web3, the different worlds and you know we're always talking about mass adoption I'm curious to hear your thoughts about that.

Ismael: 8:00

Yeah. So the beautiful thing about zero-knowledge proofs is that it is an emerging research area that's accelerating faster than I think people anticipate. So I started Lagrange about three years ago and when I started Lagrange, the concept of a virtual machine that used zero-knowledge proofs, like a general-purpose VM that you could prove, was outlandish. The concept of being able to prove an ML model was like fanciful. It was like you know. It was Narnia and then ZKML.

Ismael: 8:30

The reality in the last three years is that the cost of generating zero-null proofs has decreased by an order of magnitude plus per year. We're about 1,000 plus X cheaper to generate a proof today than we were when I started with Edge. And so now ZKML is actually usable for some applications, but I wouldn't go and say you could prove an arbitrary inference in real time on an LLM in ZK. But you can use it for some models and you can do it very cheaply for very simple statistical models. You'll use in financing and that was unheard of three years ago. Use in finance and that was unheard of three years ago.

Ismael: 9:05

And ZKVMs are now in a price point where we're talking about them becoming real time for proving within the next six months. So it's getting cheaper, it's getting more practical and when we talk directionally, what are we going to see next year? Two years, three years we're about an order of two magnitude away from this being usable for almost any computation in AI. And that's what we're very excited about, Because when you have the conflation of verifiability, cheaply at scale with AI, you get, at that convergence, just a very powerful ability to build these next generation applications that are truly decentralized, open and verifiable.

Joeri: 9:42

Yes, it seems that blockchain isKeyPay and then AI. It's not 1 plus 1, 3, it's 1 plus 1, 5. It can be so huge. For me, now, privacy and security are major concerns in digital identity, finance government applications. Can you explain how ZK key based redefine, redefine trust in a world that, yeah, relies actually on centralized verification?

Ismael: 10:12

yeah, there's two properties in zero knowledge proofs that people generally like to use a technology for the first property is just verifiable, which is I can take a statement and I can run the proof that if you were to execute that statement yourself, you get some results. That's great for inference, right? I can run an inference on a model and I can give you a proof and you can trust that I ran the inference correctly. We talked about that right there in deepfakes earlier. That's a very, very powerful property.

Ismael: 10:39

Another very powerful property that you're touching on right now is this concept of privacy and security and security of information.

Ismael: 10:47

So how can you ensure that, if you have an inference or you want to train a model over some private data without disclosing that underlying data, that you can do so in a way that preserves the privacy of the people who have that data?

Ismael: 11:10

So, for example, if I'm an insurance company and I want to train a model to predict premiums based on a bunch of decedent information that I have in my archives, right, I don't want to disclose to the world what this decedent information is and nor do I potentially want to develop the model in-house. I want to outsource that development to someone else, but I want them to be able to train that model over a set of private data and private inputs. And that's where you can really get some powerful properties. When you combine zero knowledge with other emerging forms of cryptography like MPC and FHE, you can have verifiable inference, verifiable training, in a multi-party context where the underlying information is kept private or with encryption over encrypted data. And when you combine ZK with other emerging primitives, you're able to start having this true remote privacy right. I can, in my phone, take a picture and I can outsource an inference over whether or not that picture contains illicit information or contained illicit media to a model that will prove that it doesn't, without ever having to disclose that model with the images.

Joeri: 12:14

That's very powerful yeah, it sounds really powerful. Now, for me as a marketer, it's also interesting. The next, the next question I will ask you because I see many Web3 projects struggling with go-to-market. So I'm wondering from your experience, what are the biggest marketing and adoption challenges when introducing new complex solutions in blockchain like yours?

Ismael: 12:40

I would say that there's a universal problem in the space that we continue to see, and that problem is users tend not to want to pay for privacy and security until they get burned by it. They are willing continuously to use the cheapest product of a given embodiment, irrespective of what the underlying properties of that might be, whether that be security, liability, centralization, et cetera. They're willing to use a centralized exchange over a decentralized exchange. They're willing to have a custodian over holding their assets in a non-custodial section, and then they lose money on that. The memory of having lost a material amount of assets through a principal agent issue generally isn't enough of a concern for them to stop that behavior.

Ismael: 13:32

Next time and this is what we universally see when people sell technology that adds safety or liveness or decentralization or, in the case of ZK, security of computation or verification of computation, trust of an AI model, they need to be able to have a mechanism to verify that the outputs of the model correspond to the correct model. They need a mechanism to keep their data private. It's a tough sell because a lot of people don't care right. They're willing to give all of their data to to facebook or to twitter or to telegram with no concern over the underlying safety. Um, and then you know they, they, they have these massive breaches and you know it's the equifax breach. Large amount of social security numbers are leaked and people get outraged for two days and they forget about it.

Joeri: 15:33

It's an interesting challenge, actually, because at the moment I'm advising a company, Secrets Vault, and they have a solution to protect a seed phrase with an image and also cryptography, and so it comes into play. But people are only worried about losing their seed phrase at the moment that they need it and they have the problem.

Ismael: 15:53

Exactly.

Joeri: 15:53

That's a bit the same thing.

Ismael: 16:09

that's a fantastic and so I'm going to give a cop-out answer that we were very, very lucky early on to have hired fantastic research leadership that has enabled me to focus on commercial functions while the company can continue to pursue deep tech innovation.

Ismael: 16:26

So our chief scientist, babis Pavlouantou, is one of the directors of the Applied Cryptography Lab at Yale and a professor at Yale and a professor at Yale. We have another professor on our team, Dimitris Pompadopoulos from HKUSD, and a number of researchers Ravon Trindasen, nikola Gai who have pursued a large amount of CK research both prior to Lagrange and at Lagrange. We pride ourselves on being one of the few companies that has a proper research function, which has given us a leg up on both the publication of novel research and the generation of novel research as a business. So, for reference of why this matters, our ZKML library that we've watched is, with our current benchmark, 700 times faster about than the closest alternative, and that's entirely because of the research team that we have been able to build Two professors, massive number of PhD researchers and that's a fundamental differentiator of the business, and I can go out and do BD and the team is able to continue to innovate.

Joeri: 17:24

That's amazing. If you can have a team like that, it takes time to build it. You are I don't know if you like soccer, but I'm Belgian. If you watch soccer team, yeah, you're training and you have already. Your team is there. You can do something, but you need the right players and you seems like you got the right player. So that's uh. I mean to hear now, let's have a look in blockchain. Everything is going so fast. Uh, but looking at the yeah, but the trends now in the markets, which are the opportunities you see in a zk technology over, let's say, the next two, three years or maybe some pitfalls, yeah, that maybe people are not paying enough attention to yeah.

Ismael: 18:05

So lagrange has made a very, very aggressive step recently into launching our zk machine learning product. I don't think anyone in the market saw that coming. We really just kind of came onto the scene one day with numbers that were three orders of magnitude faster than we're close with competitor and we've since partnered with. We've announced four, but we have about 15 companies in pipeline that we'll be announcing over the next two weeks when we think ZK machine learning is the next frontier, because AI trust and safety is such a prescient issue that a lot of companies right now are tackling with, and it's increasingly relevant in how AI is using the DeFi on chain, and so, as one of the categories I'm very, very excited about is obviously AI, agentic AI, decentralized inference, centralized training, competitive markets for training, model optimization, fine tuning All of this stuff, I think, is a tremendous market opportunity in crypto that people are hyped about and rightly hyped about, but they're also underestimating how large it will be at the limit. Then there's other stuff that don't directly relate to what the branch does, that, as an investor and as someone who cares deeply about the space, I'm very excited for. Examples of that are real world assets. This is actually Digital Assets Summit this week in New York and a lot of people are about talking to RWAs. How are you able to bring the institutions on chain? How can they bring structured products on chain? How can we have new types of debt issued on chain? I also think there's this theory of DeFi 2.0 that we're going to see emerge. There's companies like 3Jain, earnify, royco, that are experimenting with DeFi primitives that are impossible before. The new architectures and new infrastructure that will design the cycle, things like coprocessors, verifiable AI, things like ZKTLS All of that is infrastructure that enables these new DeFi groups, and this DeFi 2.0 revolution is going to be huge.

Ismael: 19:57

And finally, I'd say we're entering this phase of new L1s. Right, we kind of saw the L1 market really cool off in like 2022 after FDX and 2022, 2023. 2023, after FDX? I don't 2023. 2023, after FTX? I don't know. Anyway, I'm just going to need you, but cooled off after FTX. Now we're seeing a pickback. We're seeing alt SVM L1s, we're seeing new move L1s. There's a lot of interesting things being done Re-architecting execution layers, not at the roll-up stack, but at the L1 layer.

Joeri: 20:32

The time goes so fast. I also always need to think when did this happen? Oftentimes the time goes so fast it's more years ago than we think. Okay, we talked a lot about ZKP, but probably there are misconceptions in the market, larry. There are misconceptions about Web3 and crypto in general that a lot of people think directly about scams, or there may be some misconceptions about ZKPs.

Ismael: 21:02

Yeah. So I think the biggest misconception about ZKPs is that they are Ethereum-centric. Almost all of the companies building ZKP infrastructure right now are on Ethereum. They're building ZK for rollups that settle on Ethereum. They're building ZK VMs for optimistic rollups that use Ethereum. They're building ZK coprocessors to scale execution of guess what? Ethereum and EVML2s.

Ismael: 21:30

It's like this convergence of, or this overlap of, evm and ZK is like near universal right now, and so we see that like, as people are less excited about ethereum and they're looking to explore maybe some alt vms and some other kind of places to build their apps, people are assuming zk is like decreasing in prominence and no one knows it's going to happen. But the reality is zk is a universal technology that is highly applicable to any chain, any execution environment, any application, and so one of those areas that we are excited about, as I mentioned a few times, is the AI, and this is AI that doesn't have to touch Ethereum in the world. It can be selling to autonomous systems and Web2, self-driving cars, humanoid robots that need verifiability of inference. It is a very powerful technology for that reason, and the use initially to scale Ethereum, of ZK was the first application. I don't think it is long-term about that application.

Joeri: 22:31

Yeah, sorry, sometimes I say Z because I'm Flemish and it should be Z of course.

Joeri: 22:35

No, of course. No, of course. But yeah, interesting to share all these misconceptions, because there are always a lot, and with this podcast, that's what I try to do really educate people about all these new technologies, because you are using them every day. But for a lot of people, you know, I want to explain all of it Now. You've been through corporate, you've been VC, now the startup world. That's also something I would like to ask you Is there maybe one lesson from each stage of your career that you think every web3 entrepreneur should know?

Ismael: 23:12

I think the biggest thing I've learned from corporate is how to interact with large organizations. I think a lot of people in crypto tend to over index on the speed of a large corporation. They start thinking that large corporations are going to come and take over parts of the market and that they're going to be able to drive massive business value and buy massive amounts of assets because there's this huge conglomerate threat. Well, if BlackRock has $20 trillion in their management, it must buy $1 trillion of Bitcoin, right? But these things move so slow we're talking five 10-year time horizons to make any type of meaningful decisions and changes, and so institutions moving into crypto is both the most exciting and least exciting thing possible. It is the most exciting thing over 10-year time horizons, the least exciting thing over one, two, three-year time horizons.

Ismael: 24:12

That's what I've learned from corporate, having been on the other side of doing these deals, and what I've learned from corporate, having been on the other side of doing these deals, and what I've learned from venture is markets and that investors and young entrepreneurs in particular think exactly the same way and have to think the same way, and a young entrepreneur who wants to start a business and wants to be successful at it has to ask themselves like where are my users coming from?

Ismael: 24:35

What am I building? Where am I going to find PMF and how do I prove that I have PMF to everyone else and I think people also as operators, they over-index on being the first to do something, and one of the most important things you learn as an investor is you don't invest in the first, invest in the best, like someone being first if they can leverage that advantage to be the best. But what's much better than being first is being best, and so you should be ready to enter a market and compete with all of the incumbents and have a decision and a strategy on why you're going to beat them. You don't need to be the first one to enter that I love that they have a few examples from history.

Joeri: 25:16

when napster came, you know like nobody talking about them anymore because then you have Spotify and other better solutions. Yeah, I think there are lots of examples about that. You should not be the first. You should be the best Netscape navigator Long time ago. But okay, yeah, maybe we are coming towards the end of our podcast episode, but for our listeners now founders, marketers, and they want to stay ahead of the curving web tree if there is one action they could take today to prepare for the future more decentralized future what would be the action that you would advise them to do?

Ismael: 25:55

that you would advise them to do. Yeah, I would say that the benefit you have if you are not an incumbent builder with a business is that you can take a beginner's mind approach to identifying where you can generate value by creating a company in crypto. The problem with crypto today is that so many of the companies are homogenous, right, everyone is trying to build something that's ostensibly like just a marginal increase and marginal performance benefit on what someone else is doing, like. There's a million roll-up stacks. There is a million DA layers and multiple DA layers. There's multiple proving layers, there's multiple sequencing layers, multiple ZKTLS layers.

Ismael: 26:38

If you're entering the space today ask yourself where do I get users from? Where does the next batch of users or next successful app come from? What does it look like? And if you can solve that, you have all the power. And if you're an outsider or you are not an incumbent building a business, your perspective on how to accomplish that is probably different than all the people who have companies today, because you're not stuck in the same thought patterns that everyone existing industry has. And if you can leverage that advantage to generate something that has real users and distribution, you'll have all the power.

Joeri: 27:13

Absolutely. And yeah, thanks for sharing that. And actually that's one of my expressions Don't follow the path that others made, you know, but don't follow the path where it goes, but make your own path. And then and then, of course, I think to do all of the other things that you mentioned. Ismael, feel free to give if I forgot to ask something. Give one last thing. No, I think I'll give one last thing. I can give one last thing.

Ismael: 27:36

I I think you were I'll give one last thing. I can give one last thing. I will say that we are at an inflection point of crypto, that the use cases that we see in crypto that actually matter, I would argue, are a barbell strategy People get too obsessed with. At the current state, where can I build something that might be marginal and useful? But in reality, there's two things that we know for a fact are useful in crypto, and that is the cross-border remittance of assets, the use of crypto as a store of value, of medium of exchange, and the completely emergent area of AI as crypto.

Ismael: 28:16

On the complete other side of the spectrum, the only way to develop a truly unstoppable AI model is to deploy it on high liveness, the centralized infrastructure that can't be shut off, and the only way to do that is crypto. So, at the complete end of the most complicated thing you can build, which is an ASI or AGI on chain, you know you have what is an asi or agi on chain. You know you have what is probably one of the strongest pms you'll ever see in the space and completely on the other side, you see one of the strongest pms in the space, the simplest thing that every blockchain can do send value, and we find very little that's durable in the middle. So I would say that's probably my takeaway I it's a very interesting network.

Joeri: 28:59

Amazing. I really love talking to you. And when blockchain and AI come together, actually I will lead a moderator panel, I think in May, here in Portugal around that. I gave a talk in Cairo about the conversions, but then on the marketing side, when Web3 and AI come together, it was such a pleasure to have you on the show, Ismael. Likewise, if people want to know more about you, they want to know more about everything you're doing in the space. Where would you like me to send them?

Ismael: 29:27

I would say go to our Twitter at LagrangeDev one word or go to our website Lagrange Dev. You can learn more about what we build, you can start working with our products and you can start seeing some of the great customers we've been serving.

Joeri: 29:41

Very good. As my listeners know, ismael, there are always show notes, always a blog article. People can go there find all the information, can read all of the takeaways after they listen to the episode. So again, thank you so much.

Ismael: 29:58

It was a pleasure to have you.

Joeri: 29:58

Thank you so much for having me. Guys, what an amazing episode. I learned so much. If you learned so much as me, be sure to share this episode with your friends, other entrepreneurs, people that are around you that could benefit from this episode. Please share it with them. Also. If you're not yet following the show, this is a really good moment to do this. If you haven't given me a review yet, these five stars would help me to reach an even bigger audience and, of course, I would like to see you back next time. Take care.