Knowledge Graph Insights

François Scharffe

As AI unleashes a host of new tools and agents to help build knowledge graphs, engineers like François Scharffe are tasked with harnessing these new capabilities and deploying them effectively.

François brings decades of AI experience to the goal he set for himself earlier this year: to automate his own job. He's making good progress, but he's also finding that there's still plenty of need for him and other humans in the workflows he's designing.

We talked about:

the upcoming 2027 edition of the Knowledge Graph Conference
his long history of working with various kinds of AI
the distinctions between knowledge engineering and software engineering
the rationales behind his choice of Pi as his coding harness
an overview of his harness set-up and knowledge engineering workflow
the benefits to knowledge graph building that AI brings
the variety of tools and agents he has built for tasks like requirements specification, data sourcing, ontology design, term extraction and re-use, entity resolution, data mappings, ETL pipelines, etc.
the evaluation processes he's using to assess the benefits of his new workflows
the ongoing need for humans in the loop, particularly for tasks like ontology creation
how he has adapted his work style for the AI era, in particular the number of tasks he now does in parallel with AI
how he manages ongoing improvements to his system - tracking decision traces and implementing introspection steps as well as devising scenarios and using "critic" agents to improve processes
the reason he downplays prompt writing and review in his operation in favor of system evaluations
his scientific approach to system evaluation, aiming for reproducibility when moving from one model to another
the focus on agent memory in the 2027 Knowledge Graph Conference program

François' bio
Francois Scharffe is a hands-on enterprise knowledge architect and semantic AI leader. He has built and stewarded production knowledge graphs across several domains, designed ontology, taxonomy, metadata, entity resolution and data integration capabilities, and connected structured knowledge systems with AI and agentic workflows. His career spans enterprise engineering, data and AI strategy consulting, product leadership, research, and community building. François is also a founder of the Knowledge Graph Conference, a leading forum for knowledge-centric AI technologies.
Connect with François online

LinkedIn
email: francois at knowledgegraph dot tech

Resources mentioned in the interview

Pi coding harness
the original xkcd "compiling" cartoon
the cartoon updated for the AI era

Video
Here’s the video version of our conversation:



Podcast intro transcript
This is the Knowledge Graph Insights podcast, episode number 58. AI unleashes a whole host of new tools and agents to help engineers build knowledge graphs. Harnessing these new capabilities and deploying them effectively is the job of the modern knowledge engineer. François Scharffe brings decades of AI experience to the goal he set for himself earlier this year: to automate his own job. He's making good progress, but he's also finding that there's still plenty of need for him and other humans in the workflows he's designing.
Interview transcript
Larry:
All right. Hi, everyone. Welcome to episode number 58 of the Knowledge Graph Insights podcast. I'm really excited to welcome back François Scharffe. He was in the first episode of the podcast series a couple of years ago with his colleague Tom Deely. They run together the Knowledge Graph Conference. But François is also a practicing in-the-trenches knowledge engineer in New York City, and we're going to talk today about some cool workflow and architectural stuff he's developed. But welcome, François. Tell the folks a little bit more about what you're up to these days.

François:
Hey, Larry. Great to be with you today. What am I up to today? A bunch of things. I mean, the number one thing is the Knowledge Graph Conference, which we're preparing for May in New York. So we're moving from Cornell Tech to Manhattan, Central Manhattan, Bryant Park. This is exciting, larger venues, more rooms, larger rooms. And we're going to open the program, the call for presentation soon. So if you're interested to discuss knowledge graphs, you have interesting projects to discuss, then yeah, keep an eye open for the call that's going to come out soon.

François:
Yeah, but I think you invited me to talk more about my other activities around knowledge graphs. As you say, I'm a knowledge engineer. I started my career in academia, and I've always come back to it at some point, either teaching or doing research. But now for several years I've been engineering, so I've been a knowledge engineer, in the weeds of the practice, and recently got pretty excited to see how AI is changing the job. So happy to discuss that if you're interested as well.

Larry:
Yeah, no. We had a quick chat. I was in New York last week, and we had a quick chat about it, and I just got unreasonably excited about what you're up to because everybody's wrestling with this stuff now, like how to incorporate LLM, generative AI and other AI tech into their engineering workflows of all kinds. But specifically for this audience, for a knowledge engineer to be doing the stuff that you're doing...

Larry:
Well, I guess, first tell me a little bit like... I guess like any software engineer these days, you feel obligated to use these tools. And then you have to harness them in some way, and that's led to this new term, I think arose within the last year, of the harness. And you're working on a very specific kind of harness specifically for knowledge engineers. Can you talk a little bit what the impetus was for that, why you decided to approach your work this way, and then maybe just a quick overview of how it works?

François:
Yeah. Yeah, yeah. So, I mean, the first thing I want to say is I'm a knowledge engineer, but also a background in general machine learning and data science, natural language processing. And I've been following the evolution of neural networks to language models and large language models for years, being more or less involved practically with them, but close to the developments there. And remember GPT-2 coming out and the poems it was writing about unicorns and finding that so amazing. But how long ago was it? I can't even remember. 2018, 2019.

Larry:
I was going to say that's about eight or 10 years. I remember.

François:
Right.

Larry:
Yeah.

François:
So when there started to be a big excitement about AI, I was originally quite skeptical, still following but not really being so enthusiastic as to its possibilities to do everything, especially in my field, knowledge engineering, that... We can come back on the details of it, but it gets pretty complex in terms of the sort of workflows that are involved. But yeah, last December there was this realization that... I think it was Claude 4.8 or GPT 4.8. Sorry.

Larry:
I think it was Claude. If I remember right, all my engineer friends got excited about Claude, I think is the one that really like, "Okay, here we go."

François:
Right. I think it was Claude. Yeah. And I was like, "All right. I think something has changed now. We passed a capability level, and let's go and jump to that." And so now being a knowledge engineer and being a software engineer, those are separate jobs. We're pretty clear about it. And I think the models and the software coding agents, like Claude Code or Codex or others, they have large prompts that makes them specialize for coding, so for coding tasks, for code reviews, for code basis maintenance, et cetera, et cetera. And that told me that those were not necessarily the best tools for building an AI that does knowledge engineering.

François:
Because one reason is that this will take a lot of context space with specific instructions for coding that are not necessarily relevant for knowledge engineering. Although there is coding involved in knowledge engineering, it may not be as complex. You will write scripts or some programs, but it will be always in a part of a certain workflow in a repeated and self-contained way compared to general software engineering. So that was one thing.

François:
And then another was we need specific tools and specific workflows for knowledge engineering. But also, I think there was part of me that likes open source and building my own things and adopting Claude Code was for me like, "This is something we have no control over. We don't even know what's in it." Now we know a little bit better, but at the time.

François:
So I was looking around and found this minimal coding harness called Pi. And there are others. There's OpenCode, but I went for Pi. It's very minimal and extensible and made for being extensible, for being extended. And so I went there and adopted it and was like, "Hey, I'm now an experienced professional, and I've been doing that job for years, and I have a lot of knowledge on what a knowledge engineer is. Why don't I try to build an artificial knowledge engineer to automate my job?" And the New York came, and the New York... I woke up, and I was like, "All right. This year, I'm going to automate my job. I'm going to make myself useless." So yeah, that was the goal with it.

François:
And yeah, harness, to go back to your question... So a harness is a minimal environment that sort of helps contain your wild LLM, your wide AI models. If you just talk to that model, interact directly with the model and asks it to do some knowledge engineering task, it will go left, it will go right, it will go where it decides to go like a wild animal, a wild horse, but not necessarily bring you from A to B in a straight line. Eventually you may reach your point, your destination,...

Jessica Talisman

When the idea of the semantic spectrum was introduced 25 years ago, it was meant to be a conceptual tool to help programmers and architects understand the relative benefits of different levels of semantic effort.

Now, a couple of decades later, Jessica Talisman has repurposed the model as a way to organize the actual process of developing enterprise knowledge infrastructure, which she shares in her new book, "Ontology Pipeline: A Framework for Knowledge Engineering."

We talked about:

her new book, "Ontology Pipeline: A Framework for Knowledge Engineering"
the inspiration she took from Obrst's original semantic spectrum
her language-first approach to ontology
her pre-RDF semantic work with standards like Dublin Core
the three "warrants" from library science that guide functional requirements specification
the origins of her book in a 2018 Department of Justice project and its conceptual development in subsequent projects
how the ontology pipeline emerged from her process of incrementally delivering artifacts in projects
the importance of teasing out and addressing assumptions that can derail ontology projects
the crucial need to meet an organization where its at in terms of its current semantic capabilities
how her education-first approach helps get enterprise buy-in for ontology projects
how using RDF in vocabulary and taxonomy work paves the way to broader ontology work
the power of working with concept graphs, in particular its benefits in AI architectures
the benefits of symbolic representations of concepts like attribution and provenance in AI systems
the dual nature of her book as both an instruction manual and an ongoing reference resource

Jessica's bio
Jessica Talisman is a Semantic Engineer and Information Architect with more than 25 years of experience building knowledge infrastructure across enterprise architecture, e-commerce content systems, digital libraries, and knowledge management.

She has led semantic architecture initiatives at Adobe, Amazon, Overstock, Pluralsight, Senzing, Vanguard, the Department of Justice and client engagements spanning the Smithsonian and the USC Shoah Foundation.

Jessica is the creator of the Ontology Pipeline™ — a framework that moves from controlled vocabularies through taxonomies, thesauri, and ontologies to fully realized knowledge graphs — and founder of Contextually LLC, her consultancy in ontology modeling, NLP integration, and knowledge graph design. She also works with The Knowledge Graph Academy, a cohort-based program training the next generation of ontologists and semantic engineers.

Jessica publishes the Substack newsletter Intentional Arrangement. Her book, Ontology Pipeline: A Framework for Building Knowledge Infrastructures(Technics Publications), arrives September 2026.
Connect with Jessica online

LinkedIn
Intentional Arrangement
The Knowledge Graph Academy
Intentional Arrangement SKOS tool

Order Jessica's book

Ontology Pipeline: A Framework for Knowledge Engineering

Video
Here’s the video version of our conversation:


Podcast intro transcript
This is the Knowledge Graph Insights podcast, episode number 57. When the idea of the semantic spectrum was introduced 25 years ago, it was meant to be a conceptual tool to help programmers and architects understand the relative benefits of different levels of semantic effort. A couple of decades later, Jessica Talisman repurposed the model as a way to organize the actual process of developing enterprise knowledge infrastructure, which she shares in her new book, "Ontology Pipeline: A Framework for Knowledge Engineering."
Interview transcript
Larry:
Hi everyone. Welcome to episode number 57 of the Knowledge Graph Insights Podcast. I am super extra delighted today to welcome to the show Jessica Talisman. Jessica, if you've been living under a rock, just in case, I will introduce her as the person. She's the owner and founder of Contextually, a knowledge graph consultancy. She's an instructor and a curriculum designer with the Knowledge Graph Academy with Tony Seale and Veronica Heimsbakk and Katariina Kari. She's also the author — and this is what we're going to talk about today — of the forthcoming book, Ontology Pipeline: A Framework for Knowledge Engineering. So welcome, Jessica. Tell the folks a little bit more about what you're up to these days.

Jessica:
Thank you. So yes, I am working as a consultant and so I'm working with clients and customers. Also, to Larry's point, teaching and developing new curriculum and new course options for Knowledge Graph Academy soon to be released. And then my book, Ontology Pipeline, is based off of a framework I developed that was a riff on the old semantic spectrum, but is a composable and decomposable way to build knowledge infrastructures. And the book will be released and available somewhere around the September 16th and 17th.

Larry:
Excellent. I'll probably drop this on either the August 31st or September 1st. So folks will have a couple of weeks. I'll put a link to the book in there. Well, that's interesting. I didn't know that this was originally a riff on the semantic spectrum. Are you talking about Obrst's original semantic spectrum and then the subsequent. Yeah, talk a little bit about that.

Jessica:
Yeah. So I felt like it was an abstraction at the time of where semantics lived, but those of us within the semantic and knowledge engineering space have always acknowledged that semantics does exist on the spectrum. And when I say semantic, I mean context and meaning type of semantics where we introduce ontologies and taxonomies and these information and knowledge artifacts. But they are varying ... You can think of it as a spectrum of maturing structures to be able to go from something where you're just simply giving context and meaning all the way to introducing formal logic into a system. So that's where we start with controlled vocabularies. That's step one. The second step is metadata schemas, which is a super important step that's most overlooked, I think. Then taxonomies, then thesauri, then ontologies and then knowledge graphs. So it's a framework that brings you through the iterative states of building a knowledge infrastructure.

Larry:
Yeah. That's so funny. I knew that it paralleled that, but I hadn't thought about that was the riff on it. Because the thing I really appreciate about your work is that you come out of this ancient tradition of library science and well, duh, you're going to take a language-first approach. And so vocabularies make a sensible starting point from that way as well. Was that just ... Not serendipity, but just like, well, duh, that's just the way the world is. You start with vocabulary and build from there.

Jessica:
Yeah. Well, and I think it really came out of working. I started my career back in the late 1990s and it was right at that precipice when RDF was being introduced and I was working on this huge Steven Spielberg project, largest visual history catalog ever. To make two to six minute segments of this video available to end users. And we had more than 11 and a half years of footage back to back. So if you imagine this huge catalog, how do you make these little clips? It was well before YouTube and clips. Discoverable. So it was very, very early proof of concept and that at the time was based entirely on language because we had to tag content, we had to have consistency.

Jessica:
So we essentially built a thesaurus library or librarian style that really emerged out of the library sciences and cataloging practices and records and those sorts of things. You can think of it as the golden records. But the idea was having this consistency because where do we define meaning as humans? And that's what it really was working towards. And thinking about how do I present something that is a repeatable process within an organization that logically makes sense that starts with humans, which is how we define things. Because ultimately when we build a knowledge infrastructure, it's about how humans define this meaning, the first step is not how machines define meaning.

Larry:
Yeah. There's so much in there. There's five podcasts just in what you just said, but I'm really curious about when you said defining meaning in that project. 30 years ago, you don't have RDF, you don't have SKOS, you have-

Jessica:
We have Dublin Core. We had Dublin Core.

Larry:
Oh, right. So you have your library tooling.

Jessica:
I had the library.

Larry:
Yeah. Tell me about, I guess maybe a quick contrast between how you did it then, just quick overview of how you use Dublin Core for the ... And I assume that was key to your metadata strategy and everything that drove that. And then how would that unfold today?

Jessica:
Oh gosh. And you know what? Actually, that's a really good point. I would say that a lot of the practices and a lot of the core tenants of those types of systems ... So using Dublin Core, which has a more mature standard for data catalogs, which is called DCAT, and we also have DPROD, which is for data products. There are these really cool W3C standards that have evolved out of the library sciences that are based on Dublin Core. Dublin Core is not ... It's so elegantly designed because the Dublin Core format first started as 15 core elements, and those core elements were like title, description, author, all of the basic metadata that you would want to apply to things with digital ecosystem. So it was that formative ... It doesn't declare anything as a class or a property. The original thing was you just had 15 elements that could be used to describe documentation in a digital space. So that one ... And I think of it as a seed, that one seed emerged into this whole family of standards that are used to describe things in digital ecosystems.

Jessica:
So for example,...

Eric Little

Knowledge graphs are complicated. They might entail multiple ontologies, several data sources, and any number of generated graphs and other context-establishing elements.

Eric Little came up with the idea of "super domains" to understand and manage these important but ephemeral pieces of an enterprise's semantic knowledge architecture. A super domain might establish context that holds for just a few seconds for an ATM transaction or for several years for a clinical trial.

We talked about:

his recent move from Accenture to Knowledge3 as Chief Data Officer
his diverse background in the consulting and enterprise worlds
super domains, the conceptual framework he has developed to deal with generated graphs, reification, and other knowledge graph elements
how super domains capture context
some examples of super domains
how super domains can help you deal with temporary, ephemeral, and stochastic data
how his philosophy background led to his early exploration of the concept of context graphs
the crucial distinction that needs to be made between epistemology and ontology in semantic practice
the role of phenomenology in his conception and practice of ontology
his take on the current state of semantic technology practice
how generative AI has brought the importance of metadata to the fore and highlights the need for deterministic capabilities in AI architectures
the dearth of tooling in the semantic space and the need to deliver "semantics at the speed of AI"

Eric's bio
Eric Little, PhD is Chief Data Officer at Knowledge3. He previously was Industry Innovation Principal Director & Head of Semantic Strategy & AI for Global Assets at Accenture. He received a dual Ph.D. in Philosophy and Cognitive Science in 2002 from the University at Buffalo, State University of New York. His Post-Doctoral Fellowship at the University at Buffalo’s Department of Industrial Engineering (2002-2004) focused on developing ontologies for multisource information fusion applications. He has worked in academia as a professor in several fields at several universities, as well as held multiple management & C-level positions in the software development industry across several different industry verticals.

Having such a diverse background spanning academia & industry over the years provides Eric with a very unique set of experiences in the software development space that cuts across numerous disciplines and business verticals. After receiving his PhD and subsequently doing his post-doc, Eric has held various academic positions including Assistant Professor of Doctoral Studies in Health Education & Health Policy and founder of The Center for Ontology & Interdisciplinary Studies at D’Youville College. During this time he also started and ran his own consulting company which landed several high-profile customers across industries, including healthcare, insurance, oil & gas, and medtech.

He is a world-recognized expert in semantic technologies, data fusion applications, data modeling, analytics, and AI. He has numerous professional publications in these areas, has been featured in industry publications, and is a well-known speaker at conferences around the globe. Before working at Accenture, Eric co-founded and was CEO of LeapAnalysis, the world’s first fully virtualized semantic search & analytics data science engine, which was named the #3 Most Innovative Data Science Company In The World by Fast Company Magazine in early 2021. He also was named Most Innovative CEO by Global CEO Magazine. He brings this knowledge and his passion for innovation everywhere he goes, developing new technologies and furthering the growth and success of his client base. Eric has a very simple goal in life – just change the world by making things no one has seen or thought about before.

Eric is married and lives on the barrier island in Indialantic FL with his wife Jodi and their 2 cane corsos, Lemmy & Eddie. His daughters Gabrielle & Claudia are recent grads from University of WI Law School & Florida State University School of Business, respectively.

For his personal life, Eric is a former semi-pro musician and is an avid guitar player & collector. He currently has a small digital studio at his house where he still writes and records songs (fun fact: his previous band, Satori, from the late 80’s-early 90’s is listed in The Encyclopaedia Metallum). He is also a motorcycle enthusiast and likes to tinker on his vintage bikes when time allows.
Connect with Eric online

LinkedIn
Knowledge3

Photo of "super domains" slide in Eric Little's 2026 KGC presentation
Video
Here’s the video version of our conversation:


Podcast intro transcript
This is the Knowledge Graph Insights podcast, episode number 56. A knowledge graph is a complicated thing. It might entail multiple ontologies, several data sources, and any number of generated graphs and other context-establishing elements. Eric Little came up with the idea of "super domains" to understand and manage these important but ephemeral pieces of an enterprise's semantic knowledge architecture — context that might hold for as little as a few seconds for an ATM transaction or for several years for a clinical trial.
Interview transcript
Larry:
Hi, everyone. Welcome to episode number 56 of the Knowledge Graph Insights Podcast. I am really delighted today to welcome to the show, Eric Little. Eric has most recently become the Chief Data Officer at Knowledge3. And welcome, Eric. Tell the folks a little bit more about what you're up to and your transition into this new role.

Eric:
Yeah, thanks, Larry. It's been a wild ride. It's been about a month now. Yeah, so I left my position at Accenture where I was leading a lot of our strategy practice around semantics for AI. But I happened to get back in touch with an old colleague of mine, Tom Plasterer, who I'm sure you and everybody knows. So Tom and I have had a long relationship. I mean, ironically enough, it's kind of funny, we're both actually from Wisconsin.

Larry:
Oh.

Eric:
So we're both Packers fans. He was born in Madison and grew up there. I was born in Green Bay and grew up there. So we come from kind of a similar Midwest background. We both root for the same football team. We both got into semantic scientologies. We've both worked a bunch in pharmaceuticals and life sciences, so it's kind of strange. We always refer to each other as sort of brothers from another mother kind of thing. And so Tom approached me about the company and it was a really great conversation and a good idea.

Eric:
And it really got me thinking because it was very similar to a company I had co-founded and was CEO of before called LeapAnalysis, which was a data virtualization company where we made Fast Company's number three most innovative data science company in the world in something like 2021, I think. But I went to Accenture. I've been there for about the last five years. Accenture was great, really good company to work for.

Eric:
Massive company as you know. And so now I've gone from something that's almost 800,000 people back to something that's like six people. So a little bit getting my sea legs there, but it's been really great. I sort of jumped in, started working with some of our clients already, and we're pushing the envelope on some of this stuff that we're doing in semantics, especially around things like virtualization, federation, and really serving up a semantic backbone for things like AI.

Larry:
Very cool. It's like what everybody's kind of doing now, stuff in that area. There's so much demand for this. It's really gratifying to see it. But I'm really curious about for you as a practicing ontologist and consultant, you were doing, as I recall, you were doing more kind of standard, not standard, but industry stuff and more like... Whereas knowing Tom and you and K3, are you focused on pharma and the life sciences in this new practice?

Eric:
We are currently, and a lot of that is because we're small and we want to be focused. So we both have a really strong background. Tom, probably stronger than mine even in this, as well as our CTO, Ivan, Ivan Stankov is another guy who comes from the AstraZeneca background. So those guys have been really, really deep into pharma. I had been a consultant in pharmaceuticals. I've been a consultant in medical device. I've also though been a consultant in everything like oil and gas, finance, consumer packaged goods.

Eric:
I spent a lot of my career in the defense industry. So I used to be one of these guys with a top secret SCI clearance doing a lot of three-letter agency work and stuff like that, building ontologies around threats and stuff for data fusion applications. But Tom and I had this talk and we decided, "Hey, look, we both have a pretty good Rolodex in the space. It's a good place for us to get started. We have a lot of deep subject matter expertise and knowledge in the space, so it makes a lot of sense."

Eric:
We're also both really, really into fair data, making data both findable, accessible, interoperable, and reusable. And those standards are really prevalent in the life sciences space. I mean, if you look around life sciences, there's something like 1,100 plus ontologies to grab in the life sciences space, whereas you go to finance and you have basically FIBO, and you go to something like supply chain and you've got one or two. Right? And you've always just got a couple of papers on things.

Eric:
But so I think it's sort of by design, but also through a product of necessity. But that doesn't mean we need to stay there. We do have desires to move eventually as we grow out of life sciences. But Knowledge3, as a technology, it's not bound to the life sciences at all. I mean, we could apply this to virtually anything, but you do need good heavyweight semantics when you are doing this kind of stuff in life sciences....

Dougal Watt

Most conversations about taming generative AI focus on guardrails bolted on after the fact. Dougal Watt takes a different approach: generate the meaning first, as an ontology, and let everything else — agents, APIs, knowledge graphs — flow from that.

His standing-room-only workshop on AI-augmented ontology creation at the Knowledge Graph Conference drew both business executives and semantic engineers, highlighting the broad interest in the need for accurate, trustworthy AI.

We talked about:

his work at the company he founded, Graph Research Labs and their meaning-first approach
his very popular workshop on AI safety and ontologies and guardrails at the Knowledge Graph Conference
the implications in AI architectures of the need to balance time, risk, and regulation
the five main approaches that organizations are using to comply with regulations:

guardrails
multi-agent refinement
governance frameworks
sandboxing of agents
ontology grounding


how his Semantic Agent Harness delivers accuracy improvements better than even the best graph RAG systems
how they build decision tracing into their framework
the need to keep a human in the loop across the ontology-building process, as well as the crucial role of human judgement in automation workflows
how LLMs can accelerate the task of integrating data in various organization silos
how agentic modeling permits ongoing consistency checks on competency questions
the need for more ontologists in AI work, but also tooling that he has created to facilitate ontology work
how good ontology methods give organizations accuracy that is tailored to their unique knowledge

Dougal's bio
Dougal Watt is the CEO and Co-Founder of Graph Research Labs, inventor of multiple patents and patents pending covering ontology-driven declarative generation, governed AI agents, and automated enterprise stack creation. Previously IBM Chief Technologist and a global expert in information architecture. Open Group Distinguished Architect. IBM Certified Enterprise & Information Architect. TOGAF Certified. 30+ years across four continents building and fixing systems for some of the world's most demanding organisations. International speaker, most recently speaking about AI Safety and Guardrails at the Knowledge Graph Conference 2026 in New York.
Connect with Dougal online

LinkedIn
Graph Research Labs

Video
Here’s the video version of our conversation:


Podcast intro transcript
This is the Knowledge Graph Insights podcast, episode number 55. Generative AI gives ontologists powerful new capabilities that can help automate workflows and build and maintain ontologies. But as with any powerful technology you need wise guidance and sound methods to get the most out the tool. That's what Dougal Watt does. He has developed an approach that leverages the power of LLMs to accelerate ontology development while at the same time guiding every step of the process with human judgement and discretion.
Interview transcript

Larry:
Hi, everyone. Welcome to episode number 55 of the Knowledge Graph Insights podcast. I am really delighted today to welcome to the show Dougal Watt. Dougal is the CEO and founder of Graph Research Labs in New Zealand. He's also the former chief technologist for IBM in New Zealand before he started his company. So welcome, Dougal. Tell the folks a little bit more about what you're up to these days.

Dougal:
Thanks, Larry, and thanks for having me on. Great to be here. Well, what I'm doing these days, I started Graph Research Labs to explore a really exciting intersection between information architecture, which has been part of my practice for a long time, information architecture, graph, and AI, and in particular using ontologies and graph to solve some of the really major problems that I've seen throughout my career. Those sort of problems they ... What we've seen over time is that most enterprise IT, it just becomes really focused on the past. You see layer upon layer of systems, often over decades. Each gets laid onto the last. And then organizations, they just end up spending more time, more effort to maintain that past than they are to build the future. And unfortunately, AI is making that worse, not better. So again, organizations are layering agents onto that same siloed set of foundations, and then they wonder why data's not accurate or governance just doesn't keep up and governance evaporates.

Larry:
Yeah-

Dougal:
So what we're doing, we flip that around at GRL. We focus on generating the meaning first. So instead of the meaning being an afterthought, we start from meaning and then we move into the data and then we generate that software stack. So the idea is that you shift to modeling your business once as an ontology, and then our declarative engine generates that whole stack for you. So you can generate the things you need, governed agents, APIs, apps, knowledge graphs, MCP servers, data products. You can generate all that in minutes. So that's been the focus.

Larry:
Very cool. And a lot of that ... There's so much in there. One, I love that it's messy and AI is not making it any better, but everybody I hear in Silicon Valley says, "No, we can just fix that." But that was one of the points of your talk at ... Oh, quick background. We met at the Knowledge Graph Conference last month where you did this great presentation on ... Was it AI augmented or LLM augmented ontology creation? I forget the exact title, but a really interesting presentation with a really well-developed workflow around how to use the capabilities of LLMs productively, but with really tight guardrails, human interventions, and things like that. Can you talk a little bit about ... Well, the other thing about that workshop is it was packed. I somehow got there early enough to get a seat, which I was grateful for. What motivated you to put that workshop together and why do you think it was so packed?

Dougal:
What motivated me is I saw this extreme focus on AI and then people needing to understand ... Well, I guess I'll take it back a bit. I think what was interesting to me is that it was so packed, and there was such a strong lineup of speakers at the conference in general, but I was really genuinely surprised how much interest there was in that workshop. And what we covered, we covered a lot of ground, but it's fundamentally about AI safety and ontologies and guardrails. So that was the focus. And I've never actually had a session before where you have to open up an overflow room and then people are sitting on the floor. It's kind of crazy. For a technical workshop, that's just really rare.

Dougal:
And I think this is one of the really interesting points is that a large number of people that were there, it might've been around 50-50, were senior executives, vice presidents, and their senior technical staff. And to me, that's a really strong signal that what's important to people is this AI safety and ontology issue and it's being driven from the top of the organization. And that's exactly what you want to see.

Dougal:
And I think as to why there's that focus on the ontologies, well, everyone's tried AI and they've tried different approaches, and they're finding that getting that into production is a real challenge because accuracy's not high enough. So if you're aiming for greater than 90% accuracy, whatever the use case is, you need an approach like ontology. So ontology seems to be the solution and people are looking for that solution, they're looking for tools, but obviously they're using AI, they're using LLMs. How do those two come together? How do they meet in the middle? How do we get value from this amazing new technology and it's just a technology? But how do we harness it, couple it, make it safe, and be able to drive the kind of value that we need in our organization so it can move into production and be accurate?

Larry:
Yeah, I think it's interesting. I remember even three years ago at the Knowledge Graph Conference that the budding, the looming talks about neuro-symbolic AI, hybrid AI, the combination of the two has been there. And it's really interesting that it took three years to get to matter-of-fact, like, yeah, this is clearly the place. And it's really gratifying to hear that there were business people in there as well, because it's not hard to get us excited about that.

Dougal:
We love it. Yeah.

Larry:
Yeah, executives. But one of the things that came up subsequently to that talk is one, that need for safety and guardrails and methods of taming these beasts is that there's this kind of disconnect between the insane acceleration and pace of change in the technology and the regulatory environment that drives a lot of the requirements for these things. Can you talk a little bit about that?

Dougal:
Yeah, we do a lot of work in that regulatory space, and I think the more I think about it, I think we're kind of at a really strange point in history. So we're sort of caught, like you say, this conundrum. It's between time, risk, and regulation. And so as society speeds up, the risk grows larger and then regulations are expanding exponentially. So again, the conundrum, that's making decisions slower. So in a sense, there's a impression, there's that disconnect between politics and regulators and technology instead of people actually working together to make life better and to harness this technology in the right way.

Dougal:
So the way I think of it as time speeds up, risk concentrates. And you can think of lots of examples of this. Trading algorithms can move a financial position in microseconds and it once took days to build that position. Or like money laundering, that chain can now move through accounts before the first alert fires off and people know about it. So what that says is that speed without accuracy, that's not progress,...

Yann Le Franc

The emerging field of neurosymbolic AI — combining LLMs and knowledge graphs in hybrid AI architectures — is new to a lot of people.

For Yann Le Franc, it's the story of his career — from his academic days as a computational neuroscientist to his current work creating enterprise systems that ground LLMs in ontology-backed knowledge graphs.

Over his extensive research and consulting career, Yann has honed his ontology design process, which he's now sharing in a platform that guides others in applying those time-tested ontology practices.

We talked about:

his work on the European Open Science Cloud, the LUMEN and GRAPHIA projects, and his KG and AI consultancy
his academic evolution from cellular biology to neuroscience and then to computational neuroscience, neural network research, and neuroinformatics — and how that led to his discovery of the power of ontologies
his current work to connect LLMs and ontologies
a presentation on ontology-backed AI that he delivered at the Cultive Ta Data (Cultivate Your Data) conference earlier this year
the refreshing emergence of the word ontology as a buzzword (as opposed to its "bad word" status only a couple of years ago)
how he applies his research and innovation work to improve enterprise information systems
his work to advance the practices of ontology design and knowledge graph building
the inspiration he has taken from Robert Stevens to approach ontology development in a more aligned (non-siloed) way
his ensuing adoption of the Linked Open Term methodology, an agile approach to ontology, and his broader work to reconnect "semantic silos"
his journey into proper philosophically grounded ontology practice, and the notable absence of a canonical source of best practices
the work at the LUMEN to develop a collaborative platform that guides practitioners to the tools available to each step in the ontology process
how an enterprise ontology is the "skeleton of your whole information system, a representation of your company" — and how it accelerates BI and discovery and retrieval
how much he's enjoying the return to neuroscience that comes with his work on neurosymbolic AI systems

Yann's bio
Yann Le Franc, PhD is the CEO and Scientific Director of e‐Science Data Factory S.A.S.U. Created in 2014, e-Science Data Factory is a French Innovation Center, aiming at proposing innovative solutions to transform data as key assets for industry and public organizations making their use of AI more efficient. The company leverages its participation into European Research Infrastructure projects to transfer knowledge and innovation to industry and public organizations.

Yann Le Franc has a PhD in Neurosciences and Pharmacology in 2004. After a postdoctoral experience in the US, he worked on data management projects for Neurosciences in the context of the International Neuroinformatics Coordinating Facility (INCF) where he developed a strong expertise in ontology design and semantic web technologies. He then contributed to several Research Infrastructure projects aiming at building EOSC, the European Open Science Cloud (EUDAT, EOSC-Hub,…) as an expert on Semantic Web, ontology design and FAIR Principles.

He is co‐chairman of the Research Data Alliance Vocabulary and Semantic Service Interest Group and the FAIR Mappings Working Group. He is the former co-chair of the FAIR Digital Object Forum Semantic Group and actively contributed to the EOSC Semantic Interoperability Task Force. He has been spearheading the FAIRification and standardization of semantic artefacts (i.e. ontologies, controlled vocabularies, …) and mappings/crosswalks in the context of FAIRsFAIR, OntoCommons and FAIR Impact projects. Through e-Science Data Factory, he is one of the co-founders of the Knowledge Graph Alliance. For the last 2 years and a half, Yann has been the Head of the EUDAT Secretariat, a pan-European e-Infrastructure, providing data management services, storage and computing services to European researchers. He is leading the creation of the EUDAT Node, one of the first nodes of the EOSC Federation.
Connect with Yann online

LinkedIn
email: ylefranc at esciencefactory dot com

Resources mentioned in this interview

European Open Science Cloud
LUMEN, GRAPHIA, and OPERAS
EUDAT

Video
Here’s the video version of our conversation:


Podcast intro transcript
This is the Knowledge Graph Insights podcast, episode number 54. One of the most interesting developments in AI architectures is the emergence of hybrid systems that integrate neural-network based LLMs with symbolic AI like knowledge graphs. For Yann Le Franc, this is very familiar terrain. He began his career in the neural network world as a computational neuroscientist but then quickly discovered the power of ontologies during a post-doc. He's delighted to see his interests re-converge in the current hybrid-AI era.
Interview transcript

Larry:
Hi everyone. Welcome to episode number 54 of the Knowledge Graph Insights podcast. I am really delighted today to welcome to the show Yann Le Franc. Yann is, as you can tell from his name, he's based in France. He's in Montpellier. He's the CEO and founder at e‐Science Data Factory. He also heads the EUDAT Secretariat. He's the co-chair of the Vocabulary and Semantic Serviced Interest Group at the Research Data Alliance Association and he co-chairs the Fair Mapping Project. I have no idea how he found the time to talk with me with all that going on, but welcome, Yann. Tell the folks a little bit more about what you're up to these days.

Yann:
Yeah, thank you, Larry, for having me on the podcast. Well, as you've said, a lot of things are going on. We are working a lot on European research project to help build the European Open Science Cloud. And of course, we specialize on semantic web services, knowledge graphs, and the connection with AI. And in some of these projects, so the two projects that are currently going on are called LUMEN, and the other one is GRAPHIA. In LUMEN, we are developing a platform for industrializing ontology development and in GRAPHIA we're building a federation of knowledge graph for social science and humanities.

Larry:
Yeah, that's actually how we met. There was a conference in Zagreb last fall and put together by, I guess OPERAS, the overarching org, I guess that-

Yann:
The coordinator, yes.

Larry:
Yeah. And one of the first things I want to talk about is that I think it was there we talked about your journey from, you started as a neuroscientist and here you are this world-class ontologist. How did that transition happen?

Yann:
Yes, indeed. I started as a cellular biologist to be correct that specialized in neuroscience. And then I get my finger hooked into computational neuroscience. I was fascinated by the idea that you can simulate brain cells in computers. So move forward and keep going on working on, how do you say, simulating small neural networks, realistic biologically grounded neural networks. And then while during that time, I generated my own data and I had to deal with the data from my colleagues and I discovered a hell of working with data generated by others and even sometimes your own data.

Yann:
And I ended up after my postdoc in the US to actually join a team in Belgium in Antwerp and work for the international neuroinformatics coordinating facility based in Sweden. And there, my job was to actually develop an ontology for computational neuroscience models. And that's how I discovered the world of ontology and realized the power of it in terms of helping you structuring your data, sharing with others with the included meaning, which means the other understands what's in the data more easily. And of course, the connection to knowledge graph, where in the end you can start interlinking all your data together in a more simpler way and a more flexible way.

Larry:
Nice. We had a close call there. We could easily have lost you to the whole neural network community with that kind of background, but I love the story of how the need to understand and work with your data drove this. That's great. But that was a little while ago. So you've seen a lot of this. You've seen how neural network stuff is informing this, the benefits of ontologies and knowledge graphs in managing science data and understanding it better. And now we're all of a sudden, fast-forward to this year and we're three and a half years into this crazy new generative AI era. What's your kind of take on the current landscape? How has it changed I guess your interests and your work and your...

Yann:
We kind of see that coming with the development of very major AI. I don't remember the name (AlphaGo), but the one that actually won against the Go champion and all these big advances. And then we had the LLMs coming up, which is interesting. It's an implementation of neural network, gigantic neural networks. It's like trying to simulate the full brain. Unfortunately, it's not a brain definitely because it's just a statistical model. So it's interesting to see the democratization of the usage of the AI, at least the LLMs. It's a bit scary because I think the naming is wrong if we should keep calling them LLMs and not AI because there's nothing intelligent in these systems. It's just statistical models that gives you some interesting outputs that can emulate intelligence, but they don't understand the meaning. And that's also why I'm happy to go back to neuroscience and neural networks because now we are working a lot also on connecting knowledge graphs and ontology to LLMs to make them smarter because they are fantastic models for language.

Yann:
I'm using them also for helping me writing text in English because I'm French, so I'm writing in Frenchlish. So it's nice to have an LLM that helps me correct some of my writing but still you realize that sometimes the answers are comp