Skip to content

WS3 Piloting Activities - Adding New AI-Based Capabilities to the LDT Toolbox

Brief Overview

This Work Strand is about experimenting with advanced AI-based services to push the boundaries of what Local Digital Twins (LDTs) can do. Pilots will test and apply cutting-edge approaches such as generative AI (GenAI4EU), virtual worlds, advanced simulation and modelling, and multi-sector services. These solutions will help cities improve critical services (energy, mobility, infrastructure, risk management) while involving citizens more actively in shaping their communities.

Key facts:

  • 3–4 pilots will be selected
  • Start of first pilots: June 2026
  • Duration: 18 months
  • Funding: 50% co-funding required
  • Second Round of open calls: February 2026
  • Third Round of open calls: May 2026

👉 The third and last round of open calls is now CLOSED. Join our Stakeholder Forum to follow the latest updates of the community. Access the pilot journey to know more about our pilots.

What is the goal?

  • Pilot AI-driven use cases that go beyond current LDT capabilities.
  • Test novel applications using generative AI, simulation, and immersive technologies.
  • Strengthen citizen participation through co-creation and democratic engagement.
  • Ensure that both advanced and less advanced communities can benefit from new AI tools.

Who Are We Looking For?

We invite applications from:

  • Municipalities, groups of municipalities, or regions, syndicates which already have a public service (whichever) and want to experiment with AI-enhanced LDT services to improve public services and resilience.
  • Cities, municipalities, or regions.
  • Technology innovators and solution providers in areas such as AI, generative AI, Citiverse/virtual worlds, simulation, and modelling, who are ready to collaborate with cities and transfer their solutions into real-world pilots.

Minimum consortium composition:

  • At least 2 public entities
  • Plus 1 other partner from the following:
    → Private entity (e.g., service provider)
    → Private association (legal status)
    → Trusted third party
    → Representative of a use-case sector

WS3 Requirements Cheat Sheet

This is a cheat sheet for Work Strand 3 (Adding New Advanced AI-Based Capabilities to the LDT Toolbox). For a complete list of requirements, please refer to the specific Call for Pilots Manual.

🎯 Minimum Conditions

  • 🏛️ 2+ local/regional public authorities from 2 different eligible countries
    → At least 1 must have a digitally mature LDT (Rq1)
    → Descriptive-level capabilities + dynamic data integration
  • 🤝 At least 1 additional partner (private entity, association, trusted third party, or sector representative)
  • 🔀 1+ cross-sectoral use case that is innovative and citizen-focused (Rq3)
    → Must include 2+ AI-based services
  • 🤖 Significant integration of AI into LDT services (Rq3)
    → AI may be used upstream (data cleaning) or downstream (analysis/decision support)

📋 Describe in the Application Form

  • Existing LDT/platforms + capabilities (Rq2)
    → Include URL/screenshots, architecture diagrams, data lifecycle
  • Current data governance for each pilot site (Rq4)
    → Target governance across political • technical • legal • organisational
  • Current use of AI/XR/edge computing in the public authorities (Rq32)
    → How, for what purpose, ethical/legal considerations
  • Shared local challenge addressed by AI-based service (Rq5)
    → Justify AI added value vs. non-AI solution
  • Alignment with EU priorities and LDT4SSC objectives (Rq6)
    → Green Deal, New European Bauhaus, LDT4SSC challenges
  • Project management and coordination (Rq7)
    → Teams, collaboration, recruitment, political endorsement
  • End-user engagement strategy (Rq8)
    → End-users must test the service before replication
  • Quadruple Helix stakeholders (Rq9)
    → 3 of 4 groups required (public • private • research • civil society)
  • Broader applicability for other EU communities (Rq10)
  • EU initiatives alignment (Rq11)
    → DSSC, Gaia-X, AIoD, EDIHs, TEFs (CitCom.AI), SIMPL, LDT Toolbox
  • Advanced Digital Technologies used in services (Rq20)
    → Intended use of AI (LLM fine-tuning, new model, integration...)
  • AI Act compliance for AI in public services (Rq21)
    → Describe ethical risk identification, assessment, mitigation
  • Contribution to governance, efficiency, innovation (Rq22)
  • Socio-economic and environmental effects + eco-design (Rq23)
  • Sustainability strategy post-pilot (Rq25)
    → Risks & mitigation (political, social, technical, operational, business, legal)
    → Plans for at least 1 year beyond project

🏗️ Build during the project

  • Each public authority must implement its own LDT instance (≥2 instances) (Rq13)
  • Each LDT must provide management access to all 7 LDT Layers (Rq13):
    1. Data Sources Layer
    2. Data Acquisition Layer
    3. Knowledge Layer (ML/AI models)
    4. Interoperability Layer
    5. Services Layer
    6. Orchestration Layer
    7. Visualisation Layer
  • Advanced capability: Predictive, Prospective, Prescriptive, or Diagnostic (Rq15)
    → LDT must be able to simulate scenarios
  • End-to-end traceability and proof of dependencies for all AI components (Rq33)
    → Track origin, usage conditions, governance of data and software dependencies

🧱 Provide as complementary material

  • 4 draft diagrams (Rq14):
    → Technical (deployment diagram, current + future)
    → Functional (activity diagram, current + future): end-to-end data and AI pipeline
    (collection → processing → training/validation/testing → deployment → monitoring)
  • Letter of Commitment with political endorsement
  • Ownership and Control Declaration (OCD)
  • Financial Form (.xlsx)
  • Ethics and Data Protection Self-Assessment
  • Contractual framework for LDT/service sustainability (Rq24)

🛠️ Engage during the project

Pilots are expected to engage with:

  • LDT4SSC methodology phases: Explore → Validate → Define → Implement (Rq12)
  • Semantic interoperability (MIM1) using open standards (Rq17)
    → e.g., NGSI-LD, LDES
  • Interoperability self-assessment: achieve score ≥3 by end (Rq18)
  • At least 5 foundational MIMs Plus (MIM0, MIM1, MIM2, MIM3, MIM6) (Rq19)
  • LDT4SSC Assets and Services Repository for asset sharing (Rq27)
  • LDT Toolbox Marketplace for code/models/algorithms (Rq27)
  • Open repositories for data models (Rq28)
  • Up-to-date documentation for LDT4SSC consortium (Rq31)

📦 LDT Toolbox Marketplace Requirements (Rq27)

Unless using the LDT AI Notebook, AI models must meet:

  • Deployable KServe package (Kubernetes via KServe, with InferenceService YAML)
  • Accessible model artefact location (resolvable storageUri)
  • Standard inference endpoint (V2-style API at /v2/models/{model_name}/infer)
  • Documented inputs/outputs and runtime (framework/runtime, version)

✏️ Specify upon application

  • Data, assets, services to be shared, sectors involved, providers and beneficiaries (Rq16)
  • MIMs Plus (MIM0–MIM8) the project will engage with (Rq19)
    → Current and planned compliance level (Initial, Partial, Full)
  • Main expected assets to be produced (Rq26)
    → See Annex 2.9 for list of potential assets
  • Deployment approach (hosted, SaaS, on-premise) (Rq14)
  • Replicability, Transferability, Scalability measures (Rq30)
    → Describe how assets can be transferred to at least one additional context
  • Equivalent open-source solution, if proprietary components used (Rq29)
  • Citiverse components/XR usage, if intended (Rq5)

⭐ Recommendations

  • Assess LORDIMAS maturity (Rc1)
    → Pilot Lead: Digitally Optimised
    → Others: ≥ Moderate
  • Include 3+ public authorities for better replicability (Rc2)
  • Pursue alignment with LDT Toolbox (Rc3)
    → Recommended tools:
    • EU LDT AI Notebook for algorithm/model creation
    • EU LDT Federated Learning for decentralised training
    • EU LDT Data Modeller for synthetic data generation
  • Consider federation with WS1 pilots using SIMPL GA Agent (Rc4)
  • Use open-source components and share enhancements (Rc5)
  • Include IP and exploitation rights in consortium agreement (Rc6)
  • Technically establish (Rc7):
    → DCAT data catalogue
    → Data management system (JSON-LD, RDF, NGSI-LD)
    → IAM (OAuth2, OpenID Connect, Verifiable Credentials)
    → ODRL-based data policy
  • Use MIT or Apache open licence (Rc8)
  • Record baseline data for Cost-Benefit Analysis (Rc9)
  • Assess eco-design maturity (General Policy Framework for Ecodesign) (Rc10)
    → At least 30 highest-priority criteria

💰 Financial Rules

  • Maximum grant per third-party: €500,000
  • Maximum cumulative grant per consortium: €1,000,000
  • Co-funding: 50% of total pilot costs from applicants' own resources
  • Indirect costs: 7% flat rate of direct costs

🧠 AI-Specific Highlights

Please refer to our specific AI-Guidance page.

Aspect Requirement
AI Integration Significant, either upstream (data) or downstream (analysis)
AI Services At least 2 AI-based services per use case
Added Value Must justify AI advantage over non-AI solution
Advanced Capability Predictive, Prospective, Prescriptive, or Diagnostic
Traceability End-to-end for all AI components (Rq33)
AI Act Full compliance required
Ethical Risks Must describe identification, assessment, mitigation
Toolbox Publishing KServe-deployable packages with documented APIs