Assembled with existing tools.
Open WebUI provides the document and chat interface. Ollama runs the model locally. This is an integration project built around a useful workflow.
CASE STUDY / HAVENLOCAL DOCUMENT AI
Haven
A local document assistant for Dutch-speaking teams. Built around a simple question: can useful AI work with the paperwork already on the desk?
Section 4.2 · Maintenance
Relevant passage → local model → a reference you can check.
If the document does not cover it,
say the answer was not
found.
THE STARTING POINT
Finding a clause in a long specification should not require reading the whole thing again. I wanted someone to ask a normal question in Dutch and get an answer grounded in the documents.
Haven assembles Open WebUI, Ollama and a Gemma model into a local setup. My work is in connecting those pieces, configuring the document workflow and testing what happens when the answer is missing.
HOW IT FITS TOGETHER
The question, the supporting passage and the answer belong in one visible chain.
Start with a plain-language question about the supplied paperwork.
DUTCH QUESTIONSRetrieve relevant text from the document collection before generating an answer.
DOCUMENT RETRIEVALUse the retrieved material, or say the answer was not found.
LOCAL MODELINSIDE THE BUILD
Open WebUI provides the document and chat interface. Ollama runs the model locally. This is an integration project built around a useful workflow.
I also ask questions the documents cannot answer. Refusing to invent a clause is part of what I test in the pilot.
The current setup runs on my own PC. It has not been installed at a client, and there is no production rollout to claim.
A useful answer
needs a source.
WHAT I TOOK FROM IT
This project made me focus on the answer a person can check. Model output only becomes useful document assistance when its limits are visible.
NEXT PROJECT / WARDOFLIX