The models are not the bottleneck. The knowledge is.
An enterprise runs on rules, relationships, constraints, and institutional judgment — most of which live in documents, in legacy systems, or in someone's head. A model asked to make a decision in that environment is not reasoning; it is guessing fluently. The gap between the two is not closed by a larger model. It is closed by representing the knowledge in a form a system can query, validate, and cite.
That is what my research addresses: ontologies and knowledge graphs that survive real governance requirements, retrieval that is grounded in structure rather than similarity, and agent runtimes where every conclusion carries the evidence that produced it. OpenFoundry is where those ideas get built and tested.
The work is only interesting to me if it holds up in a plant.
A pattern I have not implemented is a hypothesis, not a contribution. Everything I publish has been built and run against a real system first — which is also the fastest way to find out that an elegant idea does not survive contact with a plant.
Most enterprise AI failures are not model failures. They are knowledge failures: the rules, relationships, and constraints an answer depends on were never represented anywhere a system could read them. Fix the structure and much smaller models start working.
In operations, an answer without provenance is worse than no answer — it carries the authority of a system with none of the accountability. Every design I work on treats the evidence chain as a first-class requirement, not a logging feature.
The interesting question is not how much can be automated. It is where the handoff belongs, what triggers it, and how the person on the other side is given enough context to act. That boundary is a design problem, not a policy afterthought.
My work spans industrial operations and applied AI research. Nine years of enterprise-scale transformation programmes across Europe and North America — consultant, then manager, then head of a digital transformation consulting practice — designing and delivering systems for manufacturing execution, prefabricated construction, and process industries. Environments where a design decision shows up as a bottleneck on a real line, and where the documentation is always three revisions behind the equipment.
That background is why my research is shaped the way it is. Ontology governance matters because organisations reorganise. Provenance matters because someone eventually asks why. File-based knowledge graphs exist because a graph database is one more system a plant team has to be convinced to run.
I publish what I learn, and I'm glad to hear from anyone working the same problems.
Commercial and product strategy for an Industry 5.0 software business, including the research and technology alliances behind EU-funded innovation programmes.
The self-directed programme this site documents: ontology-based enterprise AI, knowledge graphs, and next-generation operations systems — every pattern built against a real use case before it is written up.
Teaching Industry 5.0 on the Digital Transformation Executive Master — the convergence of AI, advanced technologies, and human-centric operating models.
Owned the frameworks the firm's digital consulting portfolio ran on, and led R&D programmes with universities and technology providers that put emerging technology into live operating environments.
Progressed to Manager, leading a team of eight across multi-country programmes in manufacturing, retail, healthcare, and technology. This is where the plant-floor intuition came from.
Research collaborations, industrial pilots, teaching, or just a good argument about ontologies — all welcome.