Secure AI in Action at Nexus Luxembourg 2025

One database. One secure AI-ready foundation. One future.

That was the core message of our session… and it really resonated.

---

Now mix that with vectorization (embeddings) and RAG (Retrieval-Augmented Generation) and you quickly get into risky territory if data leaves its secure home.

The traditional approach often looks like this:

1. Export data from the database
2. Push it into an external vector store or AI service
3. Try to re-create security rules somewhere else

This introduces more systems, more attack surface, and more inconsistent security models. At Nexus, many people told us the same story: the AI stack is fragmented, complex, and expensive to secure.

Our answer was different:

What if the database could be your vector store, your AI orchestration engine, and your security boundary – all at once?

---

Our Session: Secure AI Workflows Inside Oracle Database 23ai

And on top of all that, we used:

• Select AI – to speed up development and make it easier to build smart behaviors
• Oracle APEX – as the low-code front end where doctors, pharmacists, or analysts can actually use the AI securely

Everything ran inside the database.
No data movement. No external vector store. No copy-paste security.

---

The Healthcare Demo: RAG With Real Security

In the live demo (I’ll embed a short video recap here in the post 🎥), we walked through a concrete scenario:

• A healthcare professional logs into an Oracle APEX app connected to an Oracle Database 23ai backend.
• They ask a question that requires combining:
  • patient-related information,
  • internal medical guidelines,
  • and contextual knowledge stored as documents and vectors.
• The system performs a RAG pipeline:
  1. Uses AI Vector Search to find the most relevant chunks of information.
  2. Orchestrates the flow via DBMS_VECTOR_CHAIN.
  3. Ensures that Real Application Security rules are applied at every step:
     • A doctor, a pharmacist, and an admin do not see the same data.
  4. Produces a GenAI-powered answer with the supporting evidence and proper traceability.

The key point:
Even the embeddings (vectors) are subject to the same RAS policies as the original data.

That means:

• No “shadow” copies of sensitive data in external services.
• No duplicated security logic.
• No additional tools just to secure the AI layer.

As Thomas explains beautifully in his article on securing vectorized data with RAS in Oracle Database 23ai, this is where database-native AI becomes a real game changer for regulated industries like healthcare.

(I’ll link his article here in the blog so you can dive deeper into the technical details.)

---

The APEX Layer: Where Security Meets User Experience

Of course, all of this would stay theoretical if it lived only at database level.

That’s where Oracle APEX came in.

We built an application that:

• Authenticates different healthcare roles
• Calls the RAG pipeline inside the database
• Visualizes:
  • the AI-generated answer,
  • the source documents used,
  • and the underlying security constraints in action

The message we wanted to send was clear:

You don’t need 10 different platforms to build secure, AI-powered apps.
You can do it with one converged database and a low-code front end.

And based on the reactions after the session… this really clicked.

---

Conversations at Nexus: Fragmentation vs. Convergence

Outside of the session, Nexus Luxembourg was all about conversations.

Over coffee and in the expo area, we heard the same pattern from many organisations:

• Multiple databases
• Different security models
• Separate AI services
• Integrations everywhere

It works… but it’s fragile and costly. Every time you add a new AI use case, you also add more complexity.

That’s why our message —
“One database. One secure AI-ready foundation. One future.” —
landed so strongly with both business and technical audiences.

Business leaders care about risk, cost, and speed.
Technical teams care about architecture, security, and maintainability.

A converged platform gives both sides what they need.

---

Personally: What This Event Meant to Me

On a personal level, Nexus Luxembourg was special.

• I got to represent Oracle and show work that I truly believe can make AI safer and more useful.
• I had the chance to collaborate closely with Thomas, whose expertise on RAS and security is at the heart of this story.
• I felt the strength of having strategy and field experience aligned – from product vision to real demos.

The Luxembourg Oracle office also made us feel at home (it’s a small office, so I think we met almost everyone 😄).

For me, this event was another reminder that:

The real power of AI in the enterprise is not just having a model.
It’s about integrating AI securely into existing systems and workflows – without breaking trust.

---

What’s Next?

This is just the beginning for us:

• We’ll continue evolving this healthcare use case.
• We’re exploring how the same architecture can apply to other regulated industries (finance, public sector, etc.).
• And we’ll keep sharing what we learn – in conferences, webinars, and yes, here on this blog.

If you’re curious about:

• Securing vectorized data and RAG workflows in Oracle Database 23ai
• Using RAS, AI Vector Search, DBMS_VECTOR_CHAIN, OML, Select AI, and APEX together
• Or just want to understand how to simplify your AI architecture

…stay tuned. More posts (and more diagrams and demos!) are coming.

And if you were at Nexus Luxembourg and we met there:
thank you for the conversations, the questions, and the energy.