Custom AI agents and RAG, grounded in your own data
Most "AI assistant" projects fail because they hallucinate, leak data, or solve a problem nobody had. I build AI agents and retrieval (RAG) systems grounded in your own data, with citations and guardrails, on infrastructure you control. Practical AI that earns its place — not a demo.
What I build
Retrieval-augmented (RAG) assistants over your documents and databases that cite their sources; agentic document processing — extraction, classification, routing — for invoices, contracts and forms; internal AI tools wired into your real systems via n8n and APIs; and evaluation harnesses so you can measure accuracy instead of trusting a good first impression.
Accuracy and privacy first
I have deployed RAG across logistics, finance and healthcare-adjacent clients, and the lessons are consistent: ground every answer, cite the source, measure retrieval quality, and keep a human in the loop where decisions carry weight. Where privacy matters, I run models and data self-hosted and GDPR-compliant rather than shipping your documents to a third party.
How an engagement runs
A discovery call to find where AI genuinely helps (and where it does not), then a scoped build with a clear evaluation target. Most engagements start from around €3,000.
Frequently asked questions
Can AI automate invoice and document processing?
Yes, reliably — when it is built as a grounded extraction pipeline with validation and a human check on low-confidence cases, not a single prompt. I build exactly this kind of agentic document processing.
Will our data be sent to OpenAI or another provider?
Only if you want it to. For sensitive data I run self-hosted models and keep everything on infrastructure you control, which is usually the right call for GDPR and for financial or customer data.
How do you stop AI agents from hallucinating?
Grounding (RAG with citations), confidence scoring, human-in-the-loop on the decisions that matter, and an evaluation harness so accuracy is measured rather than assumed.
Let's scope it together
A short discovery call, then a clear written scope. You work directly with me — no agency layer.