As organizations increasingly turn to graph technology to better understand relationships in their data, the ecosystem around it continues to expand. To capture this evolution, Linkurious has released the Graph Landscape 2026, mapping the companies, innovations, and trends shaping the ecosystem. In parallel with this edition, we interviewed seven leading voices from the graph community, to gather their perspectives on the direction of the field, and the challenges that remain.
Our guest for the second interview in this series is Ashleigh Faith, founder of IsA DataThing, an educational YouTube channel with over five years of content focused on knowledge graphs, semantic search, machine learning, AI, and related data science topics. Her work spans multiple STEM fields and applied domains, with a consistent focus on applying complex language and data to build solutions with measurable impact.
In her conversation with Matthieu Besozzi, Head of North America at Linkurious, Ashleigh reflects on her path from SEO challenges to ontology-driven systems, challenges the idea that knowledge graphs are optional infrastructure, and explains why in an AI-saturated environment, what ultimately matters is something even more durable than data: trust.
Read the full interview transcript below and watch the complete video:
Matthieu Besozzi
So, let's talk a little bit about you.
How did you come across graph, graph technology, ontologies, RDF in your career?
Ashleigh Faith
Everybody, I feel, in our space gets into this in a different way. I did not mean to get into this, that's for sure. I was working a lot in SEO at a publishing company where they needed content to show up in search results on Google and their own platform. And as part of that, I started to explore.
I always thought, there's got to be a better way to get this content to show up. And the problem was, most people don't call it an “automobile”—because it was in STEM manufacturing that I was working in. They don't call it an “automobile”, it's a “car” or your “vehicle”. There are all these words that mean the same thing. So I actually got into ontology and graph from a vocabulary mapping perspective.
But then, quickly that transformed into doing AI where, okay, well now can we do auto classification? Can we walk a graph of all of these things and not just all the synonyms of that thing, but how those things interrelate? Because again, this was manufacturing. So if you have a car, it's got an engine, it's got tires, and those all work together in certain ways. And so, I just got very fascinated about this.
I've told this before, but David Meza, he works at NASA. I went to a conference and he was there and he was talking about this thing called a knowledge graph. At this point, I had only thought of ontologies, and there was graph, like network things. I didn't know there's a thing called “knowledge graph”, because it wasn't as popular back then. Maybe I'm aging myself when I say that. But I ran into David, he talked to me about all the things that he was working on. And that was really the spark that got me to go for my PhD, and to study this full time as my career.
So it’s just because I have a passion for it, and I think that's why I keep doing so much in this space. I love it and I love helping people get more familiar with it.
Matthieu Besozzi
Okay, cool. David is a pioneer in the field, he enabled a lot in this industry. So, what is, according to you, the biggest misconception about graph technology?
Ashleigh Faith
I once had an engineer talk to me about graphs in this way, where I said, “Hey, if we want to do this project, a knowledge graph would be a good solution for what we're encountering here.” And he told me, “Ashleigh, building a knowledge graph is like building this beautiful balcony on the side of a shack on a beach. We're not there yet. Like that's a cool thing someday, but it's not part of the foundation. It's not part of what we can do right now. It's a pipe dream kind of thing.”
And I just looked at him and I said, “I don't think you understand what this does.” And I was right. Later on, there was a huge project at that same organization where they were creating a big enterprise knowledge graph to help connect different resources across different teams, doing data quality with some of the ontology constraints and things like that, entity resolution. They started to see the value of it.
But I think that's still pretty common, where a lot of people just think a knowledge graph is a “nice to have”, not a “necessary”. Although with AI, that's obviously changing. With symbolic AI, where people are seeing a huge decrease in hallucinations, and an increase in accuracy and a decrease in risk when they introduce knowledge graph into the mix of their AI pipeline.
That has other issues though, where misconceptions are, “you need a giant graph now to do this,” because that's kind of what you needed from before. And actually when you're doing symbolic AI, you don't need a huge graph. It just needs to help where the AI falls short and when you need higher accuracy and when you want to reduce risk, which is not on everything, right?
I made a video on my YouTube channel, actually recently, on that topic. I had people coming at me with pitchforks. I was so surprised. It's like, hey, if you have a big knowledge graph, there's lots of need for that. If you have one and you want to try to use it for AI, that's also great. But for those that are starting to get into graph, and you're specifically doing it for symbolic AI, you don't always have to have a giant graph to do that.
Matthieu Besozzi
I agree, I've seen so many people actually knocking on the door saying “We have those billions of data, we need to put them into a graph,” but actually no. You should put on the graph only what needs to be in the graph.
Matthieu Besozzi
So what is the most exciting thing that happened this year in the graph? And don't say it's GraphRAG.
Ashleigh Faith
I won't say GraphRAG.
Well, I would say—maybe popular unpopular opinion—I don't ever have to write a SPARQL query again. That makes me super excited. There's lots of lovers of SPARQL. I'm not a SPARQL hater in any way. But what I'm saying is, especially when I'm trying to teach engineers how to interact with a graph, having to teach them SPARQL if you're dealing with RDF based stuff, it's tough, it's really tough.
And so now with AI being assistance here, you can have the AI write the query or you can use it to at least get the draft query for yourself. Because you still have to know that it's properly formed, it's not gonna run for 12 days, there's stuff that you still need to watch for.
So, maybe that's not to everyone else the most exciting groundbreaking thing. But I feel like there's a higher adoption of graph because people don't have as many barriers to entry. So I think that's why it's exciting.
Matthieu Besozzi
Yeah, I fully agree with you. I recently worked on a project and basically what used to take a couple of days, writing queries, fine tuning those queries, just takes now 30 minutes. So it's just, in terms of productivity, it's just awesome.
Matthieu Besozzi
What is the model that you would use for writing queries? What is the best one?
Ashleigh Faith
It really depends on what models you're using predominantly on your project or your organization. Because the more models that are out there, the worse off the model degradation is, and environmental impact and all those other things. So, I actually try to use as many of the same models as I can to all the other ones I'm doing for other projects.
So, right now, I'm looking at a lot of the Anthropic things. But I would say a lot of them do a really good job. So pick your poison, pick the one you like the most or that you're already using a lot of, and it should do the trick. So I can't really say that there's one that's better than the other.
And the other thing is—and this is just like a tidbit for anyone that is working in the AI space—don't over invest in model selection because these models are going to change in six to nine months anyways. So you're going to switch anyways. So it's okay.
Matthieu Besozzi
Very wise words. All right, so how about the future? Where do you think this graph technology is heading to?
Ashleigh Faith
Yeah, I think that there's gonna be a lot more graph adoption in areas that we used to really fight to have a place in.
One of the benefits of having a YouTube channel that's totally free, open—I make no money on the thing, I just do it to help people—is people feel very safe reaching out to me and talking to me about the things that they're struggling with. I have had so many people reaching out that are dealing with supply chain, specifically in food and areas where the manufacturing of the food gets onto the trucks kind of thing.
Maybe that's because of the other things going on in the world, but this was happening even up to two years ago. There's so many think tanks now that are focused on how to use graph specifically for shipping and manufacturing, but also in food sciences.
To see so many different people coming out of the woodwork to ask about this, yes, GraphRAG is awesome. Having symbolic AI, I know a lot of people saying like, 2026 is the year of ontology. I would fully agree with that.
So, instead of rehashing what probably a lot of other people are seeing and talking about, which I agree with. I would also say that some of the industry that's really going to jump off is being more effective at getting food from where it's being produced, where it’s being grown, wherever it is, and getting it to its final destination in a lot more effective ways. Which is going to help a lot of people, so I'm really excited to see how graph is used in that space.
Matthieu Besozzi
Now let's talk about the technology, more than use cases. Graph is an enabler of Gen AI. Graph is an enabler for users who want to make sense of complex connections in the field of whatever use case, great. But there are many problems that haven't been solved yet. What are they?
Ashleigh Faith
I think that there's patterns in graph that haven't really been fully tapped into. Certain domains just haven't done anything, or not done a lot with.
As an example, fraud detection, cybersecurity, banking have been using graph based network analysis to figure that out for a very long time. Look at the Panama Papers, right? And some of that stuff.
The thing with this is, you can use that same methodology to understand when there's false positives out there in social media, for instance. If somebody is interested in understanding across the board what are those trends that are leading to the next big thing in makeup or something, a lot of social media folks will go on and say, “Look, this thing is going viral.” And then they actually are making it viral by saying that, when actually it wasn't viral.
So that's a false positive. If you're somebody that's trying to understand how to plan the rest of your product development cycle, having false positives in your data is really bad because you're taking action on things that aren't true.
Same with research. I do a lot of things with research in my day job, and citation fraud is a really big deal right now. So scientists are out there doing really important work. And then there are other scientists that aren't doing nearly as much work and paying citation farms to literally create fake papers to cite their stuff, which is degrading all of scientific research out there.
You can use graph for citation fraud analysis. Some of these things that we see in banking or insurance or finance, and those type of areas, can actually be used in the same pattern in other areas and do a lot of good.
So, I really think that that's where the technology is going to go. Of course, it's going to keep going forward with AI and doing fact verification with knowledge graph, reduce risk with knowledge graph in the mix of AI. But transformer models, i.e. LLMs, they've had a slowdown of the giant leaps. So until the new thing outside of transformers is invented, I heard some cool things about diffusion models recently. I think that there’s going to be some gains there, but not nearly as many as we've seen thus far. I think it's going to continue, don't get me wrong, for sure.
But I think the next big thing is going to be how do you take some of these common patterns and really use them to identify things out there in the world that AI is making it easier to make fraudulent.
Because now it's not data that's the new oil. Trust is the new oil. And knowledge graph really helps you with that. If that wasn’t a sound bite, I’ve never heard one.
Matthieu Besozzi
So, let's say that I want to get into this graph technology thing. Where should I start? How do I start? What do you recommend?
Ashleigh Faith
I always say, if somebody is trying to start into graph, you have to start with a good use case. I know that sounds so bland, but there are way too many people that I talk to, they're like “Look, I built this amazing ontology and this amazing knowledge graph. And now it's great. What do I do with it?” And I'm like, “Oh, no.”
Well, a lot of your modeling decisions should be catering to what you're going to use this graph for. So if you built it, there's a lot of modeling decisions, right? Like, you can model it in the way that you, as a subject matter expert, would imagine the world, but that's not always the best way that the actual systems are going to use this thing.
So, always start out with a good use case.
And there’s two other pieces of advice on that. So, one, when we’re dealing in the graph space, and I myself am one of them, we always show stakeholders the circles and the lines, right? We always show them the graph visual. And I love a good graph visual, don’t get me wrong. But when talking to stakeholders, I have found that, at the end of the day, these are still tables underneath the hood. If you show anyone what it looks like as a table, they immediately like it better. Like, I don’t understand! It's just, not seeing the circles in the lines will actually do you some good.
So, there’s that. But, when you are trying to understand how these circles and lines document into the use case, have a real query that you want to run. Whether it's an AI being sent in with a query to gather entities for graph grounding of some sort within RAG, or this is an analytics thing that you’re sending a query across your graph to understand shortest path or fraudulent, red flags, whatever it might be, you have to do that testing, right? And do it fast.
So many people start, even if they do have a good use case, and they feel like they have to have the giant graph to facilitate it. It’s like, no, just start very small. Some of my experiments that I ran with graph, they’re maybe like 10 to 20 nodes in a graph. That’s it. You can start small and show the progress and show the promise without over-investing.
Also don't start out with tool selection. I hear that all the time, “Which tool should I start with?” I’m like, “What kind of graph are you working with?” That's pretty important to find out before you start selecting tools.
Matthieu Besozzi
Thank you. Thank you for that.
A spotlight on graph technology directly in your inbox.
