WetWijzer Companies

Logius

WetWijzer Bedrijven is a research project to develop a chatbot that provides answers to businesses' questions about laws and regulations.

Last change on 13th of June 2025, at 13:24 (CET) | Publication Standard 1.0

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

General information

Theme

Law

Begin date

2024-12

End date

2025-10

Contact information

https://www.logius.nl/contact/formulieren/algemeen-contactformulier

Link to publication website

https://www.digitaleoverheid.nl/innovatieproject/wetwijzer-bedrijven/

Responsible use

Goal and impact

WetWijzer Bedrijven is a BZK-funded innovation project that helps entrepreneurs make complex laws and regulations more accessible and understandable using an innovative prototype. The project researches and experiments with new technologies. Entrepreneurs can interactively ask questions and get answers about laws and regulations by using a combination of smart search technology and generative AI. In doing so, the prototype is entirely based on Open Source Software (OSS) and Open Language Models (OLM).

Considerations

Entrepreneurs appear to experience a significant regulatory burden and struggle to navigate the complex landscape of laws and regulations. The lack of a clear overview leads to uncertainty and possibly non-compliance. This can result in negative consequences for the entrepreneur. WetWijzer Bedrijven aims to improve this situation and offers entrepreneurs a solution. We want to increase the findability and accessibility of information on laws and regulations, focusing on the needs and questions of entrepreneurs. The prototype has a specific scope and focuses on two use cases: the Green Deal - CSRD and hospitality Rotterdam.

Human intervention

WetWijzer Bedrijven and the applied language models do not make autonomous decisions and are not part of an automated decision-making process without human intervention. Each generated response is provided with a source reference to ensure transparency and verifiability.

Risk management

The prototype is an implementation of a so-called Retrieval-Augmented Generation (RAG) architecture and is in a testing and development phase. Use is for informational purposes only. No rights can be derived from the information provided.

What the prototype does:

- Answer questions about the domain of CSRD law and hospitality for Rotterdam in the most accessible way possible only within the framework of available, public data.

- Refer to relevant legal documents and sources.

What the prototype does not do:

- Give legally binding advice or personal consultations.

- Cover topics outside the mentioned domains.

- Substitute professional legal advice.

Legal basis

The laws and regulations state that the government has an obligation to publish laws and regulations.

Elaboration on impact assessments

In addition to the IAMA test, an opinion was also requested from the DERK committee (DATA Ethics Council Chamber) of RFO. The advice is mostly positive: the prototype meets the ethical principles. The committee notes, however, that the focus is now mainly technical, leaving some concerns underexposed.

Impact assessment

Impact Assessment Mensenrechten en Algoritmes (IAMA)

Operations

Data

The sources of information used consist exclusively of official legislation from the European Union and the Netherlands. For this prototype, information from the European Green Deal, the CSRD (Corporate Sustainability Reporting Directive) and local regulations of the municipality of Rotterdam were specifically used. Additional documents from content specialists, such as from the SER, have also been included.

Technical design

The prototype is based on a Retrieval-Augmented Generation (RAG) architecture and uses several algorithms. The core of the approach is, based on the user query, to retrieve relevant information from one or more public data sources, such as laws and regulations. This information forms the basis for an answer generated using by language model.

It consists of two main components:

- Search functionality (retrieval): A search engine identifies relevant documents within the available data sources. This selection serves as input for answering the user query.

- Generative language model (generation): An open source language model (Mistral) generates an answer in natural language based on the retrieved documents driven by a system prompt. This includes references to the original sources, enhancing transparency and verifiability.

This approach increases the reliability of the generated answers and reduces the risk of so-called 'hallucinations' (erroneous or fabricated information).

External provider

No external supplier was involved in its development. It has been designed, developed, realised and made available to the Open Source community entirely under our own direction.

Link to code base

https://gitlab.com/koop/innovatielab/wetwijzer-bedrijven