The Ontology Gap

Transcript

[music] >> Just a little brief intro about myself. So, I left Google around 2018. I was been very passionate about crypto and have been building on-chain products since then. On-chain natives, DeFi trading solution, DegenScore for those of you guys remember it. Um and uh for the past couple of years the team and I have been building a product called Lighthouse, um which is basically a portfolio management for on-chain asset managers. And I'm also partner at uh Nemo Trading, which is um an on-chain uh vault curator.

So, today I'm going to take you down the um blockchain data mines with me. The team and I have spent years looking at on-chain data and despite the industry being 10 years old at this point, we still believe that the biggest unlock for what crypto can be tomorrow is really going to be in that gap between getting raw on-chain data into interpreted valuable on-chain data. And I'd like to start with a story. And um this is a an on-chain story, so this is my favorite types of story. And our protagonist here is the address on the left.

Um I'll let let's say he's a person, could be an agent for all we know. And um uh this this protagonist, this depositor, did a transaction uh with the smart contract at the block that you see. And for those of you guys who are not familiar with block time, we're looking at basically November 4th, 2025. We know that this user deposited around $4 million in USDC into this protocol. And naturally, as you can imagine, what the user does as soon as he's done doing the transaction, he goes on Etherscan, right? Which I'm guessing most of you are familiar with.

He goes to Etherscan and the page for him, let's say we can most likely imagine it's an asset manager, right? He's managing $4 million dollar He looks at Etherscan and he sees that okay, for my $4 million USDC, I got back 3.5 million ERC-20. There is no name, there is no price, there is no icon. Not particularly financially relevant. So then he does what's natural. He goes to his own wallet and he looks at the UI of the wallet and he tries to get a further context.

So those 3.5 million nameless ERC-20s now get interpreted. We get a name, we get an icon, we get a price. We don't know exactly where the price comes from, but let's assume it it's working. And so this is already a little bit more interesting, right? Financially the our depositor, our protagonist knows that okay, I have 3.5 million tokens in this usual boosted vault. There is an icon and I know that's it's worth around $4 million.

But our user wants to know more. So now he goes to a specialized DeFi app and this could be your Zerion, your DeBank, uh or the the the protocol front end as well. And now things get a little bit more interesting. So he doesn't see anymore that token before. The token gets interpreted into a position. So now this is a uh a financially relevant abstraction that it's built on top of the data we saw so far. It's a position. So now we know that this is um assets into a Morpho vault. We know that this is yield generating and we know that the USDC was the deposit asset.

So now our user starting from the from from Etherscan up to the specialized DeFi app, he has a better understanding of what's happening from his financial perspective. Although this is literally the best that we currently have. So if you're a sophisticated multi-million dollar asset manager, this is all you're going to get. And is that enough? Well, if you're familiar with DeFi and you understand that these protocols are all connected one another, if we wanted to really try to understand how this position actually works, we could try to map a sort of knowledge graph between the entities, how they're related to each other from a financial perspective. Okay, so we can see that

the our protagonist had a wallet. He owns those nameless ERC-20 we saw that represent the claim on a Morpho vault, which was allocated by a curator, in this case MV Capital, which supplies to specific lending markets. And in this particular example, one of those market was collateralized by XUSD of Stream Finance, which was already happily mentioned um on the on on stage a couple of talks ago. So, basically, we we have to see that the risk actually lives in the relationships between the entities on chain itself.

Unfortunately, as was mentioned before, um Stream Finance, on the same day of the deposit, announces basically that the protocol was collapsing and out of $200 million of TVL, half of it was roughly gone. And this impacts our depositor, the protagonist of our story, and he loses around 200k in 7 days. So, basically, what happened is that XUSD uh de-pegs, loses uh price, and the vault has socialized the losses across all the depositors. So, what we see is that the user who actually end up doing two deposits, uh end up losing around 200k in the span of 7 days.

But the interesting kicker is is that actually the user deposited after the Stream Finance collapse actually happened, and he didn't deposit once, but he deposited twice. So, what is the story uh showing? It's really showing that the data that we get on chain is really not suited and the data that we get from a lot of the the the the the the projects and front ends is still really sub par if you're really managing large quantity of money or you're trying to manage assets professionally. And so, we think that again, the the the really necessary step to go from the current self-referential state of the industry to a place where really

finance can be the on-chain finance can really be the software for all finance is really through a better understanding, a better management of data. And it's pretty odd, at least to me, that after 10 years we're still here trying to grapple ourselves with how much data we produce and yet how little of it is really actually meaningful and valuable. And specifically, if we're talking about what type of data do we need, how do we cover this gap from here to where we want to go, what type of data do we need? And I would like to talk specifically about three types of data. So, the first one is data that helps us answer what is this.

So, data that specifically references the taxonomy of an entity. So, data that lets us understand what an entity is. But we also need better data to understand how does it work. How does this protocol, how does this asset work? And that's data that needs to map how these entities relate to each other. And that's ontology, which we think is one of the most interesting challenges of the space. And ultimately, we also need data to take action. So, we need to be able to package this better data into products that can let us actually interact and use on-chain finance better. So, that's third category.

Talking about taxonomy and better data that shows what something is. I'll try to make a simple example. Um, we have two vaults here. So, one is a Steakhouse Prime USD C vault, the other one is a Nimo Prime USD C Vault. And if you look at the raw blockchain data, the two Vaults look practically identical. They're both the same ERC standard. They're both using USD C as a deposit asset. And we would exactly call both of them Vaults. But if you go a little bit deeper and we really try to understand and categorize and create a taxonomy of what each of these things is, we would see that actually the nature of the two products are radically different. From a

financial perspective, from a risk perspective, and from pretty much anything that you need to understand when you're trying to invest and access a product. You would see that the strategy that's being run is is different. So, the SteakHouse is running collateralized lending, the other one is running basically multiple multiple strategies around the options and basis trading. The manager has fundamentally different limits, is bound in a different way. The Oracle and the way the NAV is structured is is calculated is completely different. And ultimately the liquidity, so how can I access, how can I exit, how long does it take for me to get the assets, is just completely different. So, again, we clearly see that the on-chain data is really just not enough and we need to be able to dig

deeper and call things for what they are. And again, the challenge is not that we cannot describe it. The problem is that we have to do this job every time we talk about things. Every time an analyst, an agent, an asset manager goes and look into an asset, if we're not able to have a shared taxonomy that we all agree on, then the work needs to be duplicated and done again and again. And I think it's hard to understate the importance of taxonomy. And this is very evident when we try to strike the parallelism with traditional finance, right? So, if you can imagine an analyst in Hong Kong trying to understand a company and creating a report, he then sends it to his partner here in Lisbon.

They might operate on different currency. They might disagree on the valuation of the company, but they don't need to explain to each other every time what is an asset, what is a liability. Not everything has to be redone time and time again. There is a very strong and robust taxonomy within traditional finance. Of course, traditional finance runs Balkanized. So, everyone runs on different databases and they don't have the elegance of a unified in the in the few chains in a of of a unified infrastructure layer that on-chain finance has.

However, as I was mentioning, what we are really lacking on the taxonomy side is that the on-chain taxonomy is really weak and not well understood. And sometimes even crypto natives ourself, we end up spending time trying to debate like what is a vault, what is an asset, what's the nature of a L1 native token. And yeah, and I think that ultimately this is not particularly conducive to great collaboration and moving fast and getting things done. Let us assume now that we have a good taxonomy. So, let's say that we have we we spend the time, we look into the on-chain contracts and the asset and we develop a good taxonomy

that we can agree on. Is that enough? Is that enough to run assets on chain? Is that enough to make on-chain finance go global? Well, not really. If we actually look at the example before, the the one of our protagonist, we saw that the actual risk was not hidden in the asset itself and what the asset was, but ultimately was hidden in the relationship between the assets and other smart contract that the the vault was running. So, as we can see here, trying to map how something works, so the relationship between these entities, it's just as important.

And so, how would um um an ontology for on-chain finance look like? Well, we have to take the perspective of a market participant. So, let's say an asset manager. And we can take the vault from before. So, I need to understand, let's say that I know that this is a vault. I know that it maps a specific type of uh a financial product. And I need to understand how what are the key relationships that I care about. Okay. So, if I'm going to be depositing into a vault, then I need to understand how does control look like? Who can control the vault? Who can change how the vault works?

Liquidity. How long do I have to wait to get my assets back? Pricing inputs. So, how does the vault um is uh is calculated? How's the NAV done? Uh look through fees. So, what am I paying to access the vault? Uh are there underlying fees that I'm also responsible for paying? And ultimately, what type of collateral do I have exposure to? And so, if we start mapping this relationship, now we start having a graph a knowledge graph fundamentally of how this specific entity interacts with other edges of what I really care about.

And this is finally becoming actually helpful for me to understand and underwrite a specific investment. Now, we talked about taxonomy. We also talked about ontology. So, how can we map um things? But ultimately, all of this is quite pointless if we're not able to collect this data and make it actually usable for a specific task. So, if we look at the existing type of products that we have in the crypto space, I would say that we do already have quite um strong projects that are particularly good on the taxonomy um of of things. So, for example, I would say

that CoinGecko, Token Terminal, rwa.xyz, they're doing a lot of work in trying to really understand what is what. And you'd be surprised by how much complexity there is in this work. So, for example, what is an asset? Uh is the is the bridge version of an asset part of the same asset? Is does the issuer need to be the same across different bridged asset for that to be called the same asset? All these questions require answer, and the less we agree on, the harder it gets to collaborate on. Talking about I do think also there are projects that actually have really thought through how we can actually map the relationship between entities, between project protocols, asset. And Dune DefiLlama

and and actually Debank Zerion as well should be also included. But I think a very interesting one I would like to highlight is a project called The Herd, The Echo. They're really a agent-native way to try to really understand how different contracts map to each other. And this, I would say, is the current state of the products that we have on chain from a data perspective. But again, my take here is that this is great and this is a good start, but all of this is still not enough to push us really into the next step of on chain finance as a real substrate for all world's finance. And what I'm trying to argue is that I think there is now a

new generation of products that can be built on top of this existing data that are really able to capture the whole knowledge graph. So, they can basically include a good taxonomy, a good ontology, and really make it usable for a specific use case. So, this is pretty much our life's work. So, this is what we've been doing for 2 years. So, we're basically have been trying to really capture all this data, make it make sense in a unified way, and design ultimately an application that asset managers and their agents can use in one place to really manage assets on chain. And I think that what we've

seen is that not trying to really find a way to unify the data and create a great layer for people to collaborate is not just particularly important for humans and humans that need to ultimately come together and agree on what and where to deploy, but having this really strong uh layer of data is necessary for agents to really go in and do their job. An agent doesn't cannot spend tokens every time it needs to interact, do all this type of research and due diligence, and create this ontology and knowledge graph anytime that it's interacting with a smart contract. This is not possible.

And so, trying to really bring everything into one unified app is really what we've been trying to do and build it for on-chain asset managers. We currently work with around $2.4 billion in customer assets that we track through our protocol and through our through our application. Uh mostly uh treasuries, trading teams, liquid funds, and we really see that in an age where it's getting easier and easier to just get the raw data out of the chain. You can just ask Claude to do it and it will do it very well. We think that what's really going to be important is can you really create a layer that makes sense that everyone can agree to, that your counterparty can agree to, that works well across the stack, that transactions can map neatly to protocols

and protocols can map neatly to assets and we have this very um strongly typed data layer is really becoming what's going to power the next generation of uh on-chain apps. And with that, thank you very much. If you guys have any question further later on on on DeFi and this type of things, I'm very happy to chat and you should reach out. >> [applause] [music] >> Woo!