Ram Subramanian - OpenLedger, The Blockchain Built for Specialized AI - TOKEN2049 Singapore 2025

Hey guys, uh this is Ram. I'm from Open Ledger. Open Ledger is an AI blockchain that enables users to contribute data and let them monetize their models, apps, and agents. Today I'm going to talk about why specialized AI models matter. If you take a look at it, most of the applications that we use today, Metas AI or Chat GPD or any other applications, they use generic internet models.

Models that are trained on internet data. I'm going to talk about why data sets matter, why models need to be specialized. And to give you a background, we started in the space almost 10 years back. uh our team worked with various enterprises including Sony, Walmart, Viacom, Hedera and many more. We started as a company which provided solutions for larger enterprises majorly blockchain and machine learning solutions.

And what we kind of figured out is that this needs to be much broader. It needs to be a product that can be used by the end retail as well. And that's how open started. And there are a couple of problems that you saw in the space. The first one is that today's AI is very generalized, right?

As we spoke before, AI today is predominantly trained on data that you find on internet. So it's very generalized. You can just ask questions about the weather. You can ask questions about a problem that you're facing in your personal life or you can ask questions about like what's happening in the space, right? And it's going to give you a very generic answer that you can find on the internet.

And most of these models are centralized, right? They're built on data that is centralized consumed by bunch of larger firms and they control pretty much end to end of the supply chain right from data collection to the model that's being deployed and most of these models are a black box. We really don't know what's happening inside trajecity. An enterprise using llama doesn't have any idea about what data it's being trained on. So all these models are a black box.

But this is a problem for a lot of people using in the space. It could be a machine learning developer or it could be an enterprise. This is a problem for them because they want to know how these models are built. They want to know how they can manage this entire supply chain of building an AI model. They want to know what kind of data goes in.

Right? And we are solving that by providing them an end- to-end framework where an AI model developer. It could be an enterprise or an enduser. They can have an end-to-end framework where they can collect data, manage the data and train the AI the way they want. And we use attribution to figure out what kind of data sets goes into an AI model and how that data set impacts that AI model.

It's all tracked onchain as well. And we've built a specific blockchain for this. It's a blockchain, a layer 2 that is specifically built for this particular purpose where we figure out how data is being used in an AI model. We figure out how this AI is being managed, who owns them and we also give attribution back to the data provider onchain as well. So if you provide data to the AI, if your data was useful, that's tracked onchain and then you get paid for it as well.

Why this matters? How is it different from various other platforms out there? Right? If you take a look at Kager, Hugging Face, these are platforms that has been in the space for a very long time. None of them provide any kind of data ownership.

None of them provide any kind of monetization and all of these platforms since they are centralized are not verifiable at all. Open Edger is a platform where you can own your data, have your data passed on through a model, view that onchain and you can get paid for it and you can verify that onchain anytime. And our stack is pretty simple. There is an application layer and there is a blockchain layer that relays that. So whatever happens on the application layer right from the data collection to the model development to the attribution engine all of that is provided as an app and then there is a blockchain that traces how the data is being contributed that traces what kind of data goes into a model that also traces the proof of attribution and shows you onchain verified proof that you're getting paid for the data that you provided.

There is a breakthrough that we cracked over the last um in about two years. Uh we are the first guys to crack data attribution on inference. So what does it mean? When you ask a question to charge, you get an answer, right? It is an actually an answer that is coming from how the model is processing data and it's giving you an answer based on how it processed.

So that means that we need to figure out where this data come from. Right? If an answer comes in, we can trace back to the particular data point that caused this answer and we prove it on chain. So that's our first breakthrough that we did. It's a research based in Stanford and we're the first guys to do that and we're the first guys to make it practical and move it on chain as well.

There are multiple components to our product. The first component is the data platform where you can collect data. The second component is called model factory where you can train AI models. And the third platform is basically called open lura where you can deploy these models and have people use them as well. Data nets which is the first platform lets you start your own dataet and start collecting data to build your own AI.

For example, let's say someone here is working in web3 marketing and you want to build an AI that does your work and basically uses have your users use this platform the uses AI and ask for marketing based inputs. You can basically collect data sets that is related to web3 marketing. Have users contribute these data sets to this particular platform. Think of this to be like a repository where you're collecting data from people. You're collecting knowledge from people.

And once you collect that, you could take that data repository and train an AI model with that. Like how humans we go to universities, we train ourself with the knowledge that we consume. AI models have to be same as well. They can't be generic. they have to be trained on a particular field.

So you can train this model on open ledger with the data that you collected on that particular field that you're working on. Once you train this model using our no code platform, you deploy it on open ledger and have it available along with other AI models that we have built on us. So there are people who are building models for web3 marketing. There are people who are building models for memes. We have people who are building model for solidity.

There are people who are building model for various languages. So along with them your model will be deployed as well. Think of this to be like a marketplace where other AI models are available and your model is being hosted along with them. Once these models are available, you have a chat interface similar to chat GPT where you can basically ask questions to these models and you also have like an API endpoint that can be integrated to any kind of app you're building. And if any of these models that you find useful for your own application, you can access them or as an end user how you access chat GPD, you can basically choose these model, ask your questions and have it answered for you.

And this is available on open ledger platform. So right from data collection to model development, hosting that model and letting that model be accessed by end users and app developers. All of this is available end to end in our platform. Once these models are deployed, you have them available in our platform for people to attribute and reward as well. So we have teams that are building on us.

We have Ambios which is a deepen network which checks air quality. They're building an AI model for that. We have Morpheus which is building a web3 coding model and we have many other platforms like data bootstrap which is the next Google deep mind team. They're building an AI model for like verticalized internet data. They are one of the first few guys who started building on us.

We also have enterprises working with us. So we work with Trust Wallet. We work with Walmart. Uh we recently signed up Seammens as well as one of our clients were deploying their AI models using open ledger. Why does it matter for enterprises right till now we spoke about how users can collect data and build AI models but enterprises usually have their own data.

They don't need to collect data sets. But why does open ledger matter for them? The reason why is what I spoke about in the first slide which is AI is a blockchain and AI is a black box and you really don't know how this AI really works because when you get data into an AI model you have no clarity about how chat GBD is processing it you have no clarity about how Llama is processing that all these enterprises don't want that they want to make sure these AI models are traceable they want to make sure these models are auditable as well and what we provide in open ledger is that as a platform You can basically bring your own data set. If you're enterprise, you can bring your own data set as well and train an AI model with your own data set and the model is trained can be accessed by its users and when they use them they get an answer. If something goes wrong the enterprise can trace back to which data set caused that particular answer.

They can go back to the origin of the data. They have have the entire data provenence on chain. So if something goes wrong they can go back to the data remove that particular data set retrain again and make sure that the model does not have the fall again. So we are the first guys to do that. We are the first guys to crack attribution on chain and prove that every time a model is being interacted every time a model is being inferred we can trace back to the particular data set that caused that.

And we have frameworks that are building on top of us who enable this to be much bigger than what we are building. uh these are ecosystem which are building frameworks for agents and app. This is one particular one where models are available on a platform. Now this is sort of like an engine. It's a brain behind powering humans like how it is like apps need to be powered by AI and these framework lets you basically access these models by calling through our APIs and SDKs that are available.

Uh we also provide rag along with model development. So any data that is being contributed can be converted as a rag and provided to apps as well. We have a team which is building like an AIUSD for microp payments. So anyone who is basically using rewards for rewarding data contributors they can use open token of certain microp payments they can use like AIUSD which is a USD backed by bunch of stable coins and used specifically for AI payment enablement as well. And we also have a team which is building a DEX platform where models that are being built on open ledger are tokenized and made tradable.

So if you're bootstrapping building an AI model, you need to access to funds. They can use our platform to trade these AI models as well. We currently just got overdone with our test net and wind mainet. We have about 1.1 million data contributors contributing data in our network.

more than 20 projects building on us currently and we just went live about three weeks back and we have a token that enables the entire economy within our platform. So every time someone's contributes data you get paid for that and you get paid in open tokens and it's being currently traded across Binance, Upbit and Bitum and across various other exchanges that are shown here. Thanks for your time. I'm I'm Rah. I'm one of the core contributors at Open Ledger.

We have 20 different projects building on us. Enterprises like Walmart, Trust Wallet, Seammens and many more building. And if you have any questions about what you're building and you guys are interested in AI, let me know. I'll be around. Thanks for the time.