Updates

This week I was at GITEX AI Europe in Berlin.

My personal highlight was — of course — the chance to sit in a DeLorean.
Even if this car's "future" is already 11 years behind us. I loved it!

But one basic question was everywhere and it remains the biggest challenge: how do we pragmatically bring these latest AI tools into real enterprise contexts — where data, processes, governance and measurable business value actually matter?

That, to me, is the interesting part.

I believe that the future of AI in enterprises will depend on the ability to connect technology with real problems, real data and real decisions.

An exciting journey into the future — and this time, entirely without a flux capacitor.

#GITEX #AI #KnowledgeGraphs

Vibe Coding Experiment #2

Meet Ontofelia 🦉 — a self-hosted AI-agent gateway whose long-term memory is a governed knowledge graph with owl reasoner.

Most agents "remember" by dumping text into a vector store and hoping the right chunk comes back. Ontofelia stores what it learns as first-class, validated RDF facts in an embedded triplestore (Oxigraph) with an OWL reasoner — and every fact carries provenance: who said it, when, from which message.

• It forgets completely when you tell it to. In our pilot, retrieval agents leaked "forgotten" facts 31–62% of the time. Ontofelia: 0%.
• It reliably flags contradictions instead of silently overwriting them.
• Memory is auditable facts, not opaque embeddings.

I'm not claiming it beats vector-RAG everywhere — a strong modern LLM with good RAG matches it on plain recall. The measurable, defensible edge is governed truth-maintenance: clean forgetting and contradiction handling. That's exactly the regime where "trustworthy AI" gets tested.

In the spirit of my recent posts: Ontofelia was vibe-coded with Claude (Opus 4.8), Codex and Gemini. It's a v0.0.1 research preview — rough edges included.

Try it, break it, tell me what's missing:
👉 github.com/semantification-org/Ontofelia

#SemanticWeb #KnowledgeGraphs #AgenticAI #OpenSource #Semantification

I added graph visualizations and an ontology modelling view on top of the Fuseki web UI through vibe coding with Opus 4.8.

If you work with RDF and the Semantic Web, you probably know Apache Jena Fuseki. It is one of the default tools people reach for when they need a reliable SPARQL endpoint quickly — for experiments, demos, or real setups.

My relationship with the Fuseki web UI has always been a bit mixed. It does the job, but I've often wished it had proper graph visualization built in. Building that myself used to feel out of reach — the codebase is not something you casually understand in an afternoon.

So I tried it with Claude Opus 4.8 as my agentic developer. And surprisingly, it worked: two interactive graph views and an ontology modelling panel, integrated into the existing UI flow.

What surprised me most was how it smoothly integrated the new features into the existing web UI. And I needed just a few prompts. It felt like having a supporting developer.

Extending mature open source solutions toward your own needs is becoming easier than ever.

That feels like one of the most important shifts in open source: not replacing contributors, but lowering the activation energy for people with domain expertise to contribute meaningfully and generate customizations of well-known solutions quickly.

#SemanticWeb #KnowledgeGraphs #JenaFuseki #OpenSource #AIAssistedCoding

Yesterday, I attended Data Unplugged (d:u) in Berlin. The floor said it best: "Follow the data → your way to AI."

My main takeaway: an AI agent is like a brilliant new hire on their first day — fully trained, highly motivated, yet completely unaware of how your company operates.

There's a lot of excitement around AI agents. But a model lacking your organization's context is that new hire before onboarding: intelligent, eager — but ungrounded. And the onboarding is semantics!

It was a recurring theme all day — from semantic layers to data products and data democratization. The market is converging on meaning as a foundational element: ontologies, knowledge graphs, and a layer machines can actually reason over.

One line stayed with me: the autonomous enterprise is earned through data discipline.

Trustworthy AI isn't the result of simply enhancing a model. It's earned by giving your data shared, machine-readable meaning. That's the essence of Semantification.

#DataUnplugged #KnowledgeGraphs #SemanticLayer

Exactly 20 years ago today, on 06.06.06, I started my thesis at the Distributed Artificial Intelligence Lab at Technische Universität Berlin: "Agent-based policy system for network assisted mobility management."

Back then, I was building ontologies so that software agents could share a common language — and collaborate across systems. The ideas around agent-oriented software development were powerful. But in many ways, they were ahead of their time and the ecosystem was not ready yet.

Two decades later, things are different. Semantic Web standards are mature. Knowledge graphs are used at scale. And agentic AI is creating a new need for systems that do not just generate text, but can understand context, use tools, coordinate actions, and operate on shared semantics.

For me, this feels like a circle closing. What was once mostly research-driven is now becoming reality.

Over the last decade at Bayer and Bosch, I have seen first-hand how hard — and how important — it is to connect data, knowledge, systems, and people across large organizations.

I believe the next wave of AI will need more than prompts and APIs. It will need semantics. And that is why I am starting something new.

I have many ideas around Semantic Web, Knowledge Graphs and Agentic AI — and I will share more soon. The third decade of my career is starting with excitement and gratitude. I am very much looking forward to new collaborations, new conversations, and to building what has now become possible.

#SemanticWeb #KnowledgeGraphs #AgenticAI #Semantification