Struck

Municipality of Amsterdam

AI explorer for spatial rules in digital environment plans. Helps to quickly access, search and summarise specific rules within the digital Environment Plan for a specific location.

Last change on 12th of May 2026, at 9:28 (CET) | Publication Standard 1.0

• Publication category: Impactful algorithms
• Impact assessment: Field not filled in.
• Status: In use

General information

Theme

Field not filled in.

Begin date

Field not filled in.

Contact information

[e-mail]

Link to publication website

https://www.struck.eu

Responsible use

Goal and impact

Manually searching through complex, layered digital environmental plans in the Digital System for the Environment Act (DSO) is a time-consuming process for employees. Due to the layered legal structure, it is often a challenge for non-specialists to quickly and accurately identify which specific rules are in force on a particular parcel of land.

Struck explore environmental plans aims to unlock this information quickly and in understandable language. The system allows users to ask questions in natural language about the use and building possibilities at a specific location. The algorithm analyses the relevant plan rules and summarises them directly, eliminating the need to manually navigate through hundreds of legal rules.

The impact may be that the complex structure of the new Environment Act becomes more accessible to a wider group of employees. This could lead to significantly shorter turnaround times for preliminary consultations and faster, higher-quality information at the counter for both citizens and businesses.

Considerations

The commitment is justified because it significantly increases the transparency and accessibility of complex legislation. The benefit of faster and understandable information outweighs the risk of interpretation errors, provided the human expert retains final control through the source references provided. The sources can be read in full and even a highlight reference is made to the specific section on which the AI bases the answer.

Human intervention

There is a 'Human-in-the-loop'. The AI's interpretation serves as support. The employee checks the accuracy of the unlocked information against the provided source references to the official Environment Plan.

Risk management

Risk management around Struck explore environment plans is aimed at ensuring the accuracy of the information accessed and facilitating expert verification by the user.

1. Managing data misinterpretation and accuracy: The risk of misinterpretation due to incomplete source information in the DSO is overcome because the system is designed as a tool for subject professionals. The algorithm unlocks the relevant plan rules including direct source references to the original texts. This enables the professional user to validate the completeness and context of the information independently. It is assumed that the user, based on their own subject matter expertise and the sources presented, will themselves identify when information is missing from the source text or when a situation requires further manual interpretation.
2. Information security: The algorithm runs within a protected and secure Microsoft Azure environment (SOC 2-compliant). No exchange with public AI models takes place and the entered project data is not used for training general models.
3. Vulnerable groups, exclusion and profiling: These risks are not applicable. The algorithm performs strictly legal disclosure based on objective spatial rules from the environment plan. No personal data are processed that could lead to discrimination based on individual characteristics or backgrounds of applicants.
4. Implementation status: The algorithm is currently in use as a decision support tool. Its operation is periodically evaluated based on changes in the Environment Act and updates in the Digital System for the Environment Act (DSO). Changes within the DSO are processed in real time in Struck so that the correct source is always available. The algorithm is evaluated and improved weekly.

Through training from Struck and qualified third parties, users are taught how to work with Struck in order to achieve optimal results. Prepared prompts and defined workflows (checklists) standardise the output result between users.

Legal basis

https://wetten.overheid.nl/BWBR0041297

Elaboration on impact assessments

Struck analyses plots and environmental regulations, no profiling of natural persons, no special personal data, and no listing on the AP list. The legal grounds under Article 35 AVG are not met.

Operations

Data

The algorithm uses the following source data and basic registrations to arrive at an accurate assessment:

1. Environment Act & local rules (via the DSO): The current text of the Environment Act, the Building Works Decree for the Living Environment (Bbl) and the rules from the 'Brideschat'. Site-specific rules and environmental values are retrieved via the link to the Digital System for the Environment Act (DSO).
2. Basic Register of Addresses and Buildings (BAG): data on the official boundaries, surface areas and uses of existing properties on the plot in question. This forms the basis for calculating the current building density.
3. Land Registry: Data on plot boundaries and ownership relations, necessary for exact determination of the rear yard area and its relationship to public space.
4. Project-specific information: User-entered data on the proposed building initiative, such as dimensions, location and intended use of the new structure.

The algorithm only processes object data and spatial data from these sources. No personal data from basic registrations or the DSO are processed for the operation of the algorithm

Technical design

The system uses Retrieval-Augmented Generation (RAG) with a Large Language Model (LLM) via Microsoft Azure. Based on a location and a user query, the system searches the legal texts of the local Environment Plan. The AI analyses the relevant passages and generates a summary that directly answers the query, including references to the source articles.

External provider

Struck.build BV