KiM Explorer

Ministry of Infrastructure and Water Management

KiM's scientific search and chatbot.

Last change on 17th of December 2025, at 15:56 (CET) | Publication Standard 1.0

• Publication category: Other algorithms
• Impact assessment: AIIA
• Status: In development

General information

Theme

• Traffic
• Organisation and business operations

Begin date

2026-01

Contact information

https://www.kimnet.nl/service/contact

Responsible use

Goal and impact

Knowledge Institute for Mobility Policy (KiM) has a large archive of publications. This archive continues to grow. This makes it increasingly difficult to search this properly. The KiM Explorer is an AI chatbot that helps with this. With it, users can first find documents that are important for their question. Then they can chat about these documents. This makes it easier for them to get to the right knowledge. They are always directed to the source from which an answer comes.

Considerations

Now the publications can only be searched manually. This can only be done by searching the title on the website. As a result, not all correct documents are always found, or it takes effort. With the KiM Explorer, this process goes better and more efficiently.

Human intervention

For now, only a limited group of employees will have access to the KiM Explorer. They will be given clear instructions and explanations about the limitations of this chatbot. This states that they cannot use answers just like that. They must first check the answers themselves. The chatbot encourages this. A source is always given with link. The chatbot also offers to contact KiM.

Risk management

1. Only information already in the public domain is used.
2. Human intervention is encouraged (see above).
3. During the pilot, usage will be monitored. Users must also answer questionnaires.

Impact assessment

AI Impact Assessment (AIIA): https://www.kimnet.nl/documenten/2025/11/11/aiia-2.0-for-kim-explorer

Operations

Data

1. A commercial language model of OpenAI is used.
2. The publications users can search for are public reports from KiM.
3. During use, statistics are collected on usage. This is done anonymously. Personal data and search terms are not stored.

Links to data sources

KiM publications: https://www.kimnet.nl/documenten

Technical design

This chatbot is Python-based and uses the OpenAI API to answer queries with an LLM. It incorporates a proprietary RAG architecture that works in two steps.

First, only relevant documents are found and displayed (vector store search). In this, the user has to make his own selection. Then a chat session is started, using these documents as context. This makes transparent what the scope of the conversation does and does not include, and improves the quality of the answers. The language model instructions ensure that appropriate answers are given and sources are clearly stated.

Usage statistics are collected with a self-hosted instance of Umami analytics.

External provider

Internally developed

Link to code base

https://gitlab.com/kennisinstituut-voor-mobiliteitsbeleid/kim-explorer