Many organizations are planning an internal chatbot or a retrieval-based AI application. The conversation usually starts with platforms, vector databases, and vendor offers. Whether the result is useful is decided elsewhere: in the content itself—its structure, quality, currency, authority, and who is allowed to see what.

I help organizations find out what their own knowledge can realistically support before they commit to an architecture. The starting point is the material that exists, not the technology that is available.

My Perspective

The content decides, not the stack
Answer quality depends on how knowledge is structured, maintained, and authorized. Most weaknesses of AI assistants trace back to the underlying material rather than the model.

Evidence before architecture
How a body of knowledge behaves under real questions can be observed early, with representative material. A sound architecture follows from these observations—not from assumptions.

An assistant is a knowledge system
It needs ownership, maintenance, evaluation, and governance. Otherwise answer quality erodes as soon as the content changes.

How It Starts

A first engagement is usually a focused assessment of your content and your opportunities:

• what knowledge exists, where it lives, and in what condition
• which questions it can answer reliably—and which it cannot
• where an assistant creates real value, and for whom
• what it takes: content work, integration, access rights, operation

The result is a decision basis grounded in your own material and a clear next step—from implementation guidance to building and operating the system.

Confidential Content

Sensitive documents often cannot simply be sent to external AI services. I use a dedicated local AI infrastructure with current open models and industry-standard tooling. Material can be analyzed and tested in a controlled environment—and what works there transfers to larger on-premises or cloud architectures.

Situations

Organizations typically involve me when:

• an internal chatbot or knowledge assistant is planned, but the path is unclear
• a pilot delivers inconsistent or unreliable answers
• vendor proposals need an independent assessment
• knowledge is spread across document stores, wikis, and business systems
• content is confidential and cannot be passed to external services

Good work starts with a good conversation.