GRTiQ Podcast: Special Release: Matthieu Di Mercurio on The Graph, AI, and Geo

Today I am speaking with Matthieu Di Mercurio, VP of Product and Strategy at StreamingFast. In a recent blog post released on April 15, 2024, Matthieu and his team, alongside collaborators Yaniv Tal from Geo and Sam Green from Semiotic, explored the intersection of The Graph, AI, and Geo.

In this special episode, I invited Mat to discuss the inspiration behind the blog post, the insights shared by the StreamingFast team, and why the convergence of AI and The Graph represents a tangible advancement rather than mere hype. As you’re about to hear, Mat provides compelling arguments for the evolving platform nature of The Graph and its implications for the future of the protocol.

The GRTiQ Podcast owns the copyright in and to all content, including transcripts and images, of the GRTiQ Podcast, with all rights reserved, as well our right of publicity. You are free to share and/or reference the information contained herein, including show transcripts (500-word maximum) in any media articles, personal websites, in other non-commercial articles or blog posts, or on a on-commercial personal social media account, so long as you include proper attribution (i.e., “The GRTiQ Podcast”) and link back to the appropriate URL (i.e.,[episode]). We do not authorized anyone to copy any portion of the podcast content or to use the GRTiQ or GRTiQ Podcast name, image, or likeness, for any commercial purpose or use, including without limitation inclusion in any books, e-books or audiobooks, book summaries or synopses, or on any commercial websites or social media sites that either offers or promotes your products or services, or anyone else’s products or services. The content of GRTiQ Podcasts are for informational purposes only and do not constitute tax, legal, or investment advice.



We use software and some light editing to transcribe podcast episodes.  Any errors, typos, or other mistakes in the show transcripts are the responsibility of GRTiQ Podcast and not our guest(s). We review and update show notes regularly, and we appreciate suggested edits – email: iQ at GRTiQ dot COM. The GRTiQ Podcast owns the copyright in and to all content, including transcripts and images, of the GRTiQ Podcast, with all rights reserved, as well our right of publicity. You are free to share and/or reference the information contained herein, including show transcripts (500-word maximum) in any media articles, personal websites, in other non-commercial articles or blog posts, or on a on-commercial personal social media account, so long as you include proper attribution (i.e., “The GRTiQ Podcast”) and link back to the appropriate URL (i.e.,[episode]).

Nick (00:12):

Welcome to the GRTiQ Podcast. This is a special release of the podcast where I speak with Matthieu Di Mercurio, VP of Product and Strategy at StreamingFast. On April 16th, 2024, the StreamingFast team in collaboration with Yaniv Tal from Geo and Sam Green from Semiotic released a blog post, Building a Decentralized Brain with AI & Crypto, that eloquently explored the relationship between The Graph, AI and Geo. I invited Matt onto the podcast to talk about where the idea for the post came from, the insights and observations he and the StreamingFast team shared, why this theme of AI in The Graph is concrete and not just hype and much more.


And as you’re about to hear, Matt presents another compelling argument in favor of the evolving nature of The Graph as a platform and what it means for the future of the protocol. I started my discussion with Matt by asking about where the idea for the post originally came from.

Matthieu Di Mercurio (01:10):

So around the beginning of the year, we saw everyone that was posting around their prediction for what’s coming in tech in the blockchain industry in general. Got really interested, got a lot of discussions going internally at StreamingFast, and we wanted to share some of our opinions, some of the things that we’ve been working on, some of the things that we’re excited about. And so yeah, now is the right time where after digesting all that information, after having lots of interesting discussions, after building some things even, we felt ready to share a bit more of that vision with the rest of the world.

Nick (01:54):

And I always find it interesting when I do a special highlight of a post or if I find out that someone within The Graph community is writing a blog post of who the intended reader was. But who did you and the team have in mind as you wrote this blog?

Matthieu Di Mercurio (02:09):

So we had a pretty broad audience. We wanted to bring ideas that would resonate with practitioners, with people that are building AI systems, something that wasn’t read a hundred times ’cause I’ve mentioned that we read a lot of the existing blog posts. A lot of them were recycling the same type of information. So we wanted something new, even for people that are in the field, but we also wanted to make that accessible to people that are just in general interested by blockchain technology, by AI, and how those type of technologies can overlap and can help each other.

Nick (02:52):

Matt, for listeners who haven’t had a chance to read this, can you just briefly summarize what the blog post is and what you write about?

Matthieu Di Mercurio (03:00):

Yes. So we’re talking about those two fields between AI and blockchain coming together and how The Graph, StreamingFast, Geo can help in bringing that vision to life. So really we start from the fact that AI is getting more and more centralized. A few companies are building the largest models. They’re sitting on tons of GPUs, on tons of data and those two things, the compute power and the data assets that you have is what allows you to build great AI models, especially LLMs that we’ve seen over the past year. And we think that there’s a better path forward. I think we can make that type of technology accessible to more people through decentralized solution. And one avenue to take that to more people is leveraging decentralized knowledge graphs to augment the value that the LLMs can provide to their users by feeding them with up-to-date curated information.


And so, in the blog posts, we go in more details around how we see that retrieval augmented generation. So the idea you retrieve information that is relevant to your question or to the task that you’re trying to accomplish, then you use that information to prompt your LLM, and so you feed that information to a model so that it provides a more accurate, more precise answer, so kind of bridging the gap between what decentralized knowledge graphs can do and what a lot of people have been trying for the first time over the past year with LLMs, with ChatGPT and all of those tools.

Nick (05:02):

And you go into a lot of detail within the post itself about a lot of those themes. If we think about then distilling the central argument or position that the team at StreamingFast is taking here, when I read it, I saw it as A) a little bit of like we’re on the cusp of surrendering AI to centralized web2-type companies, which could be disastrous. But there’s also this other point, which is The Graph is really well positioned to feed AI’s accurate and verified data. So I guess, with all these ideas bouncing around, what was the main argument, as you see it, of the post?

Matthieu Di Mercurio (05:42):

Yeah, so really, really those two things, I think there’s a part that where people that create data should really own that data. And right now, we’re seeing that with the lawsuits between big content producers and large companies that are training models where basically they scraped the web, they got access to some data without asking, and that data ends up being leveraged for applications that it wasn’t intended for and also without having any traceability and without really rewarding the content producer. So this is, I think, a big issue and something that we’re starting to talk about and starting to open a solution space for this, like an avenue for fixing that broken model where anyone can steal the data, there is no traceability of all this. So I think that’s one part.


And the other part is getting into a bit more concretely how The Graph can help. So really thinking about The Graph’s positioning as an indexing technology. And in general, The Graph has a central role to play in the future of AI. We all understand that all of those AI models are only as good as the data that you’re feeding into the model. So if we can extract better, more refined, if we can be more efficient at extracting that data, then the models will be much better. They can also be trained in a way that doesn’t require huge funding access to a fleet of GPUs and yeah, like a way where a lot more people would be able to build those models and leverage them eventually. So I think in the aspect of pre-processing and data management, The Graph is already in an amazing position to support this and has a lot of the building blocks that are required to get there.


But I think there’s a part that’s even more exciting to this, when you bring in the idea of the world of data services. So beyond just indexing, having services that are accessible on the network, and that can be run by Indexer and that anyone can apply. And the reason I’m mentioning this is because beyond just pure text generation, which is amazing, we’ve all used ChatGPT, Gemini, like any combination of those tools, they are limited to outputting texts. And the next level is building agents, so tooling that can actually interact with the world and can call APIs, not just give you a response in text, but also actually make a decision for you or like call it an external service. And where The Graph can play a role here is that if it can become the platform where all of those services run.


And so now you have that idea where The Graph helps you to collect data, to train your model or to make that data accessible to the model, but then it is the platform on top of which you can build agents so that those models can interact with the world and not only give you recommendation, but also do things for you. And Semiotic has built an agent like this where you’re starting by chatting, they get data from Substreams, so you can run analysis on the blockchain data, you can have a trend of transactions of token price of whatever you want and then based on the analysis that you’ve done helped by the LLM and the data that it has accessible, you can make a transaction through the agent.


So in text, you can ask for that request and it’s going to call an external API. So now it’s not just a chatbot that will tell you a few things, but it retrieves data, it posts data somewhere else. It’s really something that can bring the level of automation to the next level that can make us as humans a lot more efficient because now I have one UI in which I can do all of these things, whereas I would have opened Excel, do my analysis on my side, and then go on a DEX and do some transactions. All of that UX can become a lot more slick, a lot simpler, and I can see how the world of data services can be the foundation, the platform on which this type of application is built in the future.

Nick (10:52):

That’s an amazing vision for The Graph and long-time listeners of the podcast know that this idea that The Graph is a platform upon which people build has been spoken of in a lot of prior Episodes. Again, I think listeners of the podcast are familiar with The Graph, they’re familiar with AI. I’ve done a lot of Episodes and some panel discussions on the topic, but then we introduce, in this post, something about Geo, and you’ve referenced it a little bit here in some of the things you’ve shared so far, but how does Geo fit into this platform vision of The Graph?

Matthieu Di Mercurio (11:24):

So going back a little bit more to the idea of knowledge graphs. So on top of pure content, like freeform text and natural language, which it is a great foundation, but it’s really hard to parse, it’s hard to parse for humans because it takes you a lot of time, but it’s also still hard even for LLMs that are really good at this, but that data is all unstructured. So you’re assuming that the model will be able to understand from all of that text its context and how things relate to each other. Geo is a way to create that knowledge graph and connect information between different pieces of content. So you can publish content, you can connect it to different concept, different ideas, people, and the links that are being made, those relations can be curated by people. So really you have that idea of creating a graph of, well, I was about to say, a graph of knowledge, which is really what it is doing, right, like taking this idea of raw data and making it into higher-level concepts and bringing it together.


And the important aspect I think that Geo is bringing is that this is done in a decentralized way where everyone can contribute. You can’t imagine one person writing Wikipedia. You can’t imagine one person organizing the world’s information. This has to be done as a collective. This has to be validated by different people, and that’s really what Geo is helping with. And once you have that, I think that opens the gates for more collaboration, more creativity because by connecting ideas, that’s how new ideas come up. And so I think, yeah, it opens a lot more applications on top of that knowledge that has been aggregated.

Nick (13:31):

Well, as is mentioned in the post itself, Yaniv Tal, the founder and CEO over at Geo contributed to this post, as did Sam Green over at Semiotic, two names that you’ve already mentioned there. And so I think I want to ask you this question about, clearly you have to do some research, you have to do some collaborations with the Geo team, Semiotic. You’re working internally with the StreamingFast team. As you’re doing this research, did you learn anything new about The Graph or AI or even the way The Graph and AI could converge more in the future?

Matthieu Di Mercurio (14:02):

Well, that vision for The Graph being a platform on top of which AI agents can be built, I’ve been able to refine this a lot by writing this blog post, by going deeper into that topic and understanding really what this would look like. Look at other applications in different industries that are doing something maybe a bit similar, maybe some inspirations of platform plays, maybe some that are not even in the AI space, some that are parallel in the AI space, but in completely different verticals. So yeah, I got a lot deeper on that side. I got the chance to collaborate with the teams at Semiotic, at Geo, learn more about also what they were working on, what their thought was about the space and where we were going with The Graph as individual companies and as an industry and in general.


So yeah, lots of interesting insights and yeah, I think better understanding of those real applications because as we’re talking about at the beginning, there’s been a lot of talks about how AI is going to change the world, how blockchain is going to change the world, and how those two things together are going to be even better. And I think we need to navigate that well and bring that to concrete applications, concrete use cases and that’s also a big part of what we wanted to do. Sometimes when I read the [inaudible 00:15:43] blog post that was talking about something very general that was just high-level vision that had nothing that would work in the next year, for example, or in the next few years, I was a little bit, not depressed, but I wasn’t really excited by those visions and wanted to bring our piece here and make that a lot more concrete.

Nick (16:05):

So, Matt, I may be wrong about this, but it seems to me a blog post like this couldn’t have been written a year ago or before that. Is that right? Is now the right time for a blog like this, kind of talks about The Graph as a platform and explores more about Geo and AI and how it all comes together?

Matthieu Di Mercurio (16:24):

Yes. I feel like with that very rapid adoption, I would say a year ago we were at the point where GPT4 was just released. People were starting to play with it. ChatGPT was around for something like six months or so, people had an idea of where this was going. But I feel like we’ve had enough time to play with the technology to build those toy projects, small little things on the side, but then, really think about it and bring that to, as I was mentioning, real application, production use cases, something that you can put in the hands of a real user and something that’s not just like, “Hey, let’s use that new cool technology.” And I feel like we’re getting to that point.


And also a year ago, I feel like there was a lot more uncertainty of what is that technology and what are we really going to be doing with it because two years ago, no one would’ve been able to predict what happened about a year ago. But I feel like now, when I’m looking at all of the releases, all of the improvements, especially in the AI space, yes, the field is moving very fast, but it’s in a way that can be planned ahead. And so we know that retrieval-augmented generation is getting better, that models are getting cheaper, that open source models are catching up with the closed source ones. And these things are moving really fast, but they’re moving in the direction that people would have predicted maybe six months ago. And so that’s why I feel like now is a good time.


Kind of a convoluted way to answer your question, sorry, but now is the right time because I have a lot more confidence in what we’re talking about today than I would have had a year ago. A year ago, we were throwing a lot of ideas and were unsure of where this would all end.

Nick (18:31):

That’s a brilliant answer, and I appreciate it, not convoluted at all, and a great explanation as to why now is the right time to be exploring and talking about these things. You did mention this a moment ago, and I want to go back and just double click on it, but the crypto industry, and I guess you could even argue, a lot of industries throughout the world are making a lot of claims and noise about AI. And in the case of The Graph, you’re saying that there’s some real concrete use cases here that it sort of doesn’t have to fall into that trap of being all hype, all noise. What is it about The Graph that makes it so well suited to deliver on concrete value as it relates to AI?

Matthieu Di Mercurio (19:09):

Yes. So I think some of the things that we’ve talked about, the idea of being able to index blockchain technology in a way that can be compared to anything else, like much more efficiently, faster, more flexibility. All of this is at the core of everything that builds on top of AI. First, the foundational models have to be trained on high volume of data and high quality of data. So that’s just building that foundation requires technology that is aligned with what The Graph is doing. And then I think that the idea of what we’re talking about around knowledge graphs, this needs to also be in the open and a way to organize information that is not feasible in a closed environment, like organizing that information, having it curated by a community of people. This is not something that’s always going to be closed, and that you can completely trust a company to do this. And even if you trust the company to do it, there’s just so much data to curate, how are they going to do that? They won’t have an infinite supply of people to label that data, to connect information together.


New information is going to be created at a faster pace than they can consume. And so I think that’s one of the part where The Graph can support a lot. And that division for the world of data services, I think, is a really strong component that will, as I was mentioning, build that platform that people can build on top of. I feel like this is the real answer to how do we run agents on chain, how do we run those services so that they’re verifiable. And yeah, we’ve talked about some of the example applications. So with Geo, with what Semiotic has built, I think it’s just scratching the surface of everything that we can be doing. Once we start seeing those first few applications, I think a lot more developers are going to get excited about that and continue building on top of it.

Nick (21:40):

So, Matt, I think the final question I want to ask you is about the future of The Graph. So AI is a budding story. Clearly, the post that you and the team have written here tells a very interesting angle and evolution of the protocol into a platform of which it’s servicing AI. What does your blog post ultimately say about the future of The Graph? Has it changed your vision at all?

Matthieu Di Mercurio (22:04):

Well, yeah. I would say those two parts of the vision that I’m very excited about is building this knowledge graph and becoming a reference for data, for knowledge that would be accessible through the network. And that second vision of being a platform where people can build applications on top, and especially AI-driven applications where you have interactions between a lot of different services that all run on the network, all accessible through The Graph’s network and with an AI in the pilot seat like as an orchestrator of all of those integrations.


So yeah, I think that’s really the vision and the direction that we’re taking, something that we are very interested in and excited about. And I think all of this, everything that has been built in the past around indexing data, getting access to the data, making that available to Indexers and to the network, this is a very important building block towards that vision. And so to me, it’s not a shift in vision because you have to have those building blocks available to be able to start talking about more complex systems with multiple services that interact with each other. So does it completely change the vision? No, but I feel like we’ve been working in a very interesting direction and are now at an inflection point where we’re seeing that what we’ve built is really powerful, and we can continue building on top and multiply the value through those applications on top of that.

Nick (24:03):

Matthieu Di Mercurio, thank you so much for joining the GRTiQ Podcast and talking about this incredible post that you and the StreamingFast team just released with the help of Yaniv Tal over at Geo and Sam Green over at Semiotic. If listeners want to learn more about you, the team at StreamingFast, or dig a little deeper into this post, what’s the best way to get started?

Matthieu Di Mercurio (24:22):

First, thanks for having me. It has been really a great time, really, really excited to share that with everyone. So yeah, the blog post is on our medium, so You can also follow us on X at streamingfastio. And yeah, all of that information is going to be in the show notes caption, so you can find that information here too.


Please support this project
by becoming a subscriber!



DISCLOSURE: GRTIQ is not affiliated, associated, authorized, endorsed by, or in any other way connected with The Graph, or any of its subsidiaries or affiliates.  This material has been prepared for information purposes only, and it is not intended to provide, and should not be relied upon for, tax, legal, financial, or investment advice. The content for this material is developed from sources believed to be providing accurate information. The Graph token holders should do their own research regarding individual Indexers and the risks, including objectives, charges, and expenses, associated with the purchase of GRT or the delegation of GRT.