All referencesCase study · City of Dormagen
AI agents in Dormagen's immigration office
From a decree database to pre-assessing applications: how the immigration office of the City of Dormagen built out BasePeak.AI step by step. Run locally, on limited hardware, and without rebuilding the existing case management systems.
- Customer
- City of Dormagen
- Area
- Immigration office
- Hosting
- The city's own servers
- Product
- BasePeak.AI
- Press coverage
- WDR Lokalzeit Düsseldorf · 2025-11-05
Caseworkers in the immigration office handle complex individual cases every day. To do so, they need information from laws and decrees, internal guidelines, the electronic case file and existing specialist applications. Reviewing extensive files first and working out the relevant facts is particularly time-consuming.
The solution was not built in one go; it grew in stages. It started with a decree database. Then came extensive case files that had to be processed on local hardware with limited resources, followed by further legal sources from EU law to North Rhine-Westphalia state law. Today, specialised AI agents bring the case file and the legal position together.
The story in five chapters
Chapter 1 · The beginning
It started with a decree database
In immigration law, the statute alone is rarely enough. Decrees, application guidance and internal rules come on top, and anyone handling a case needs to know which of them currently apply. That knowledge was scattered, and new staff had to piece it together the hard way.
So the work began with a clearly defined goal: making relevant decrees and internal knowledge easier to access. Caseworkers ask their question in their own words; the assistant answers from the decree database and shows the source it relies on.
This first building block set the principle that applies to everything since: every answer must be verifiable against the original.
Chapter 2 · The case files
Up to 400 pages per file
An electronic immigration case file typically comprises around 50 to 400 pages. It contains previous residence permits, official letters, educational certificates, family-related information and other documents relevant to the decision.
Even the first screening of the file, the so-called file summary (Aktenspiegel), takes anywhere from 30 minutes to several hours depending on size and complexity. Only then can caseworkers bring the information together for the assessment of an application.
Processing large files in low-resource environments
For the City of Dormagen, we run BasePeak.AI on the city's own servers. No data whatsoever leaves the city network, so staff can work with personal and sensitive data without concern. At the outset, computing power there was very limited, and the system was optimised for use with small, locally run AI models.
The project in numbers
- 50–400pages
- is the typical size of an electronic immigration case file
- 30 minto hours
- is what the first manual screening of a file takes
- < 10minutes
- for a 100-page file, with locally operated models
The solution
A source-referenced file summary, produced by AI
BasePeak.AI processes the PDF export of the electronic case file using local AI models. The agent classifies and structures the relevant content and presents it as a clear file summary with references to the original file. Caseworkers can check statements in a targeted way instead of first searching through every page themselves.
What the file summary structures
Residence history
Previous residence permits and developments under residence law.
Education history
School and vocational qualifications and documented stages of training.
Criminally relevant content
Facts and proceedings documented in the file.
Family overview
Documented family members and family relationships.
These four are examples. The file summary covers further content, and which categories it captures can be configured to suit the need.
Integration
Works without a direct interface to the specialist application
Dormagen's immigration office uses OKVisa and komXwork. Their system environment offers no direct interface for retrieving the electronic case file automatically.
Caseworkers therefore export the complete file history manually as a combined PDF. This takes about ten seconds, or two clicks. The file is then handed to BasePeak.AI for analysis.
The decisive time saving is therefore not in the export, but in the review and preparation of the content that follows. For this use case, the existing specialist software does not need to be replaced or deeply integrated.
With an e-file connection With an interface to the electronic file system, the entire process up to the finished file summary would run fully automatically, without a manual export.
Export as combined PDF≈ 10 sec
First manual screening30 min to several hours
Analysis by BasePeak.AI, 100 pages< 10 min
How long a manual screening takes depends on the size and complexity of the file. The analysis time is a practical value for a 100-page file in the environment described, not a general performance promise.
How a file moves through
- Electronic case filein the specialist application
- Combined PDF2 clicks, about 10 seconds
- Local AI modelsclassify and structureThe city's own servers
- File summarywith source references
- Caseworkers and further agentsverify, research, pre-assess applications
- Decisionstays with the responsible staff
Chapter 3 · The legal position
New knowledge sources, from EU law to NRW state law
Knowing the file is not the same as knowing the legal position. A professional assessment needs both: the facts from the file and the rules they have to be measured against.
The decree database was therefore joined by further knowledge sources: federal law, administrative regulations, North Rhine-Westphalia state law and EU law. The agents draw on these sources together and show the exact reference in their answers.
This created the foundation for bringing the file and the law together.
Chapter 4 · The use cases
What the agents do in the immigration office today
The file and the legal sources have given rise to use cases that caseworkers rely on every day. The file summary from chapter 2 pulls double duty: directly for caseworkers, and as the foundation that further agents build on. A selection:
File summary
The file, structured: residence and education history, criminally relevant content, family. Every entry with its page reference.
Current decrees
Which decrees apply to a question, and what do they say? The answer draws on the decree database and shows the reference.
Legal research
Research across federal law, administrative regulations, NRW state law and EU law, with exact references instead of mere summaries.
Application pre-assessment
Build a checklist for the specific application from the legal position, then check the applicant's file against it point by point.
Application pre-assessment in detail- In testing
Decision preparation
The next step after the application pre-assessment: preparing the decision notice on the basis of the checked file and the legal position.
In detail
Application pre-assessment in two steps
A checklist from the legal position
For the residence permit applied for, the agent compiles from statutes, administrative regulations and decrees which requirements must be checked, each with a verifiable source reference.
Checking the file against the checklist
It then matches the file point by point: what is documented, and on which page? What is missing, what is still open?
Human-in-the-LoopThe AI supports information preparation and professional review. The official decision remains with the responsible staff.
Chapter 5 · Today
From the immigration office to an agent platform for the city
The first knowledge assistant in the immigration office has grown into a broader approach: specialised AI agents support a range of everyday administrative tasks, and departments configure them for their own needs.
What grew over five chapters
- The decree databaseDecrees and internal knowledge, with sources
- Large files, small hardwareFiles of up to 400 pages, processed locally
- More legal sourcesFederal, administrative regulations, NRW, EU
- The use casesFile summary to application pre-assessment
- The platformAround 20 agents for more than 300 users
- ≈ 20
- self-built AI agents in the city administration
- 300+
- users
- up to 75%
- shorter onboarding for supported tasks
Figures from the project context, as of August 2026. Significant relief has been reported for selected AI-supported activities, up to processes that took several hours shrinking to a few minutes. These values do not apply across the board to every file or every administrative procedure.
What the deployment shows
Step-by-step growth
Each chapter builds on the last: first knowledge, then the file, then the law, then the use cases.
Tangible benefit
Extensive files are opened up for caseworkers faster.
Traceability
Source references allow checking against the original documents.
Reusability
The file summary and the legal sources support several agents at once instead of being rebuilt for every use case.
Practical to deploy
Works with local models, limited resources and no direct interface to the electronic file system.
Scalable approach
Instead of a single chat application, a platform for specialised AI agents emerges across different areas of work.
First the decrees. Then the files. Now both together.
The work in the Dormagen immigration office shows how BasePeak.AI grows step by step in a restrictive local environment: from a decree database to source-referenced analysis of extensive files to agents that bring the file and the legal position together. Every stage stays verifiable against the original, and the existing case management landscape does not have to be rebuilt.
BasePeak.AI: AI agents built for the demands of real administrative work.
Let's talk
Tell us what you have in mind. We will come back with a straight assessment.