Künstliche Intelligenz für Kommunen

AI for Municipalities (Core Motivation + Problem Statements)

The video explains why municipalities should use AI, motivated by:

• Shortage of skilled workers
• Increasing the attractiveness of locations
• Closer citizen proximity / better service
• Improving professional perception
• For internal operations:
  • Reducing redundant tasks
  • Better collaboration
  • Meeting compliance requirements
  • Supporting knowledge transfer and turnover
  • Enabling innovation
  • Becoming a more attractive employer

Where AI Can Be Used (Examples Beyond the Main Focus)

• Traffic management: predict and optimize traffic flows
• Energy efficiency: optimize consumption and improve peak-time handling
• Waste management: analyze disposal and optimize waste streams
• Citizen services / digital assistants: knowledge management via chatbots
• Urban planning: population forecasts, traffic/environment modeling, recommendations
• Healthcare: prevention/treatment recommendations based on population data
• Security: predicting danger zones (e.g., police use cases)

What the Video Focuses On (Product / Solution Angle)

• Main topic: citizen services and knowledge management using digital assistants
• Example municipality portal: “Planstadt” (fictional, but used to demonstrate workflows)

Key Technical Concepts Explained

1) Limits of Standard LLMs

LLMs (large language models) are described as being trained on data (often publicly available information). For municipalities, this leads to several issues:

• Cost & rebuilding: creating/rebuilding is expensive (millions of euros referenced)
• No “unlearning” / no deletion: once trained, information can’t easily be removed
• Holistic answering: answers aren’t clearly traceable to specific topics/sources
• Hallucinations / unreliable output: the model may answer even when it lacks real information, using best-guess training data

2) “Trustworthy AI” Approach from empolis

The proposed solution emphasizes:

• Including knowledge sources with answers (citations/links to documents)
• Role-based access: answers depend on the user group and what they’re allowed to see
• Avoiding “everyone can access everything” behavior

3) Retrieval-Augmented Generation (RAG)

The described method:

1. Uses a knowledge base as the source of truth
2. Has the LLM analyze the question, check precision, and extract relevant information from the knowledge base
3. Generates an answer based on the retrieved articles/documents

4) Generative AI for Knowledge Base Enrichment

Approach for creating/expanding knowledge using internal artifacts:

• Inputs can be emails, chats, documents
• The LLM generates suggestions using structured context, with anonymization for personal data
• An editor reviews suggestions before saving them back into the knowledge base

Showcases / Tutorials Demonstrated

A) Citizen Q&A via the Municipal Portal

1. “Who is the mayor?”

• The chatbot answers using Planstadt’s knowledge context
• It provides a link to the relevant document rather than a generic response

2. Details about the mayor

• Education, marital status, career, and party affiliation are answered via linked documents
• The conversation demonstrates source-specific retrieval (different documents used for different attributes)

B) Kindergarten Registration Workflow

• The citizen asks about registering a child for kindergarten
• The bot retrieves relevant documents and presents:
  • A required topics/checklist (five topics mentioned)
  • A place availability overview:
    • Includes tables and graphics
    • Shows historical sufficiency
    • Notes limitations (availability varies by age group/region)
  • Linked “related topics,” e.g.:
    • Promotion/support for children
    • Language screening

Multilingual capability

• If content needs to be provided in French, an editor can translate a document in the system
• The example emphasizes preserving formatting/punctuation rules per language

C) Wedding Planning Guidance

• The citizen asks what to consider for a wedding
• The bot pulls correct administrative references, including:
  • Registering the marriage
  • Arranging the date and ceremony
  • Whether witnesses are required
  • Possible options/locations (e.g., “Love Bridge” with history/legend content)

D) Configuring “Standard Questions” and API-Based Configuration

• The portal provides suggested/preselected questions to reduce repeated user friction
• A REST client configures which questions appear
• Example added question: “business tax rate”
• The bot answers using extracted values and clarifies provenance:
  • Example shown: “155 out of 100”
  • Mentions extraction from a table using the model and compares relative levels (“relatively low compared to other municipalities”)

E) Generating New Knowledge Articles from Emails (Editor Workflow)

• An employee processes a naturalization request and wants a universally applicable knowledge article
• Steps shown:
  1. Import email → assign topics
  2. Remove salutation/personal data
  3. Generate a suggested structure
  4. Transfer to the editorial system
  5. Extract components like “Note,” “Problem,” “Solution steps”
  6. Save, edit if necessary, and publish into the knowledge base

Summarized Benefits Claimed by empolis

• Efficiency gains through centralized knowledge/data access
• Reduces knowledge monopolies and supports knowledge transfer (including turnover/employee departure)
• Improved quality assurance: requests handled like by the best expert
• Better service for both citizens and new/inexperienced staff
• Savings from optimized processes and better decisions
• Better information availability for citizens
• Mentions (though not fully shown) that guided dialogues, knowledge graphs/networks can support compliance with municipal regulations and enable cross-resource collaboration for innovation

Main Speakers / Sources (At End)

• Eddie Mönch Executive Vice President; responsible for non-industry direct sales and business development at empolis

• Matthias Named as “Matthias” in the narration; leads the explanation of AI basics and the solution

• empolis Company context; solution demonstrated with “empolis Body” and the “Planstadt” portal examples