Nine pilots. One reusable foundation.

Different applications, shared progress

Each pilot has a practical task and measures of success. Its role in the programme is also to test the shared runtime, interfaces and evidence needed to carry useful capabilities into another setting.

Pilot findings will inform both the common framework and possible future sector products. The descriptions below concern planned research activities.

UC1 / Synthetic populations for mobility and energy planning

Pilot lead: IRT SystemX.

Planners need credible ways to explore how populations might travel and use energy under different conditions. This pilot will combine generative models with explicit constraints to create synthetic populations for agent-based simulations. It will examine scenarios such as electric-vehicle uptake and new urban districts, assessing realism, consistency and transfer to new regions.

Evaluation focus: statistical fidelity, impossible or inconsistent agents, generation time and regional transfer.

UC2 / Trajectory understanding for urban and maritime transport

Pilot lead: Akkodis.

Movement is difficult to predict when observations are incomplete or come from different sensors. This pilot will combine video, positioning and other sensor data with traffic and maritime rules to study trajectory prediction and anomaly detection. Outputs will support human operators through risk maps and explanations, with evaluation of prediction quality, latency and rule compliance.

Evaluation focus: prediction error, anomaly detection, response time and performance in new regions.

UC3 / MANTRA: traceable answers for manufacturing

Pilot lead: Expert.ai.

Engineers and technicians often need to find a precise answer across manuals, specifications and technical rules. The Manufacturing Answer Retrieval Assistant (MANTRA) will investigate natural-language access to this material, with references to the source and checks on the answer. The pilot will evaluate retrieval and response quality while keeping technical decisions with the human expert.

Evaluation focus: answer accuracy, source traceability, coverage of in-scope questions and response time.

UC4 / HEKA: a knowledge assistant for healthcare operations

Pilot lead: Expert.ai.

Hospital staff work with protocols, procedures and knowledge distributed across many documents. HEKA will study how a source-grounded assistant can help staff retrieve relevant information, with explanations and uncertainty made visible. The planned departmental pilot will assess correctness, information-access time and user experience. Clinicians and other authorised professionals will remain responsible for decisions.

Evaluation focus: expert-rated correctness, time to access information, source references and user experience.

UC5 / CARE-S: weather-aware support for rural seniors

Pilot lead: Social Innolabs.

Weather alerts can be difficult to translate into useful daily action. CARE-S will explore clear, personalised guidance for older people in rural communities, using local weather information and simple check-ins. Recommendations will be explainable, with users retaining control and any escalation to neighbours or caregivers governed by consent. Evaluation will consider clarity, engagement and unnecessary escalations.

Evaluation focus: message clarity, check-in adherence, reported stress and false escalations.

UC6 / A cognitive companion for knowledge work

Pilot lead: ENORA Innovation.

Knowledge-intensive work involves changing tasks, unfamiliar tools and continuing learning. This pilot will investigate context-aware guidance and just-in-time learning support that adapts to user feedback. The companion will be designed around opt-in participation, clear explanations and controls to pause or disable assistance. Evaluation will examine task efficiency, useful skill-gap detection and the experience of the people using it.

Evaluation focus: task efficiency, onboarding time, skill-gap detection and user feedback.

UC7 / A digital twin for climate-smart agriculture

Pilot lead: Neuralio AI.

Agricultural planning must account for climate, land use, economics and human decisions. This pilot will develop a digital twin that combines Earth observation, climate projections and agent-based simulation to explore policy and adaptation scenarios. Planners will be able to examine assumptions and trade-offs across outcomes such as yield, income, water use and resilience, with transparent decision support.

Evaluation focus: yield and policy-scenario fidelity, adaptation across regions and clarity for planners.

UC8 / A companion for tinnitus self-management

Pilot lead: OST – Eastern Switzerland University of Applied Sciences.

This pilot will study a smartphone companion for adults living with tinnitus. It will investigate how symptom tracking, sound-based activities and coping exercises can inform explainable, personalised support over time. The evaluation will examine patient-reported outcomes, engagement and safety, with human control and a pathway to professional advice when a request falls outside the system’s scope.

Evaluation focus: patient-reported outcomes, adherence, understanding of recommendations and safety review.

UC9 / AI-assisted forecasting for coastal resilience

Pilot lead: MET Norway.

Coastal management needs detailed information about interacting pressures on marine ecosystems. Focusing on the Oslofjord, this pilot will combine operational forecasts and observations with data-driven models to examine local conditions and management scenarios. It will assess forecast accuracy and computational efficiency, with planned integration into the European Digital Twin Ocean infrastructure.

Evaluation focus: salinity and temperature forecasts, computational cost and transparent ecosystem-management scenarios.

What should transfer between pilots

The shared foundation is intended to make capabilities, evaluation methods and qualified improvements reusable. Domain data, permissions and operating procedures remain specific to the application.

Cross-pilot evaluation will test that distinction through performance, robustness, explanation quality and resource use. Reuse must be demonstrated under the conditions of the receiving system.