Research project · Proposal in preparationNot just your AI.
EVA
European Verified Agents
Not just your AI.
Your team’s AI.
What if AI did more than help individuals, and understood the team’s shared task? We want to explore this with industrial partners in production, maintenance and service.
Proposed Horizon Europe research. Funding has not been awarded.
Understand together. Decide with evidence.AI concept illustration · Not an existing product
Research in the SMC context
Making AI useful. For teams, too.
EVA (European Verified Agents) is a proposed Horizon Europe research project investigating how shared AI assistance could support teams in production, maintenance and service. We are seeking industrial partners with a concrete use case. The proposal is in preparation; funding has not been awarded. SMC’s intended contribution is knowledge capture and adoption, connecting the research with its work on AI-enabled organisational development.
For clarity: this research project is distinct from the EVA knowledge assistant on the SMC website.
The difference at a glance
From individual answers to shared understanding.
The idea: connect relevant contributions, identify contradictions and reveal missing preconditions. The pilot must establish whether this helps in practice.
Individual AI assistance
Production↕Own chat
Maintenance↕Own chat
Quality↕Own chat
Each role works with its own view. The team still has to connect the pieces.
The EVA research idea
Production
Maintenance
Quality
Connect statements and open issues. People review and decide.
Illustrative example · Packaging line
Technically ready. But cleared for production?
A setting has been corrected after a changeover. At handover, three roles report different aspects of the situation.
“Technically ready. The setting is corrected.”
“Checks are still underway. No clearance yet.”
What could EVA contribute here?
Make the missing precondition visible.
“The setting is corrected. Quality clearance is still missing. Which check is outstanding?” EVA would connect the statements to the same case and ask a focused question. The responsible person decides whether production can resume.
Concept example, not live AI or evidence of a capability already achieved.The research question
When does shared AI add more value?
We plan to compare current practice, individual AI assistance and a shared agent, using comparable tasks, evidence and core AI capabilities. The outcome is open.
Shared understanding
How does a team develop an accurate view of the task, roles and open issues?
Appropriate initiative
When do questions and proactive suggestions help, and when do they interrupt?
Evidence & trust
How do people recognise sound recommendations and detect errors?
Possible use cases
Your operations shape the pilot.
Look for recurring tasks where several roles depend on one another and experts can judge a correct outcome.
Shift handover
Carry completed checks, their rationale and open issues into the next shift.
Joint troubleshooting
Production and maintenance connect observations with tests already performed.
Service & support
Hand over a case so the next team can continue with the right context.
Knowledge & onboarding
Capture practical expertise with experts and make it usable in subsequent tasks.
Engineering changes
Connect current instructions and product variants with their implications for each role.
Demonstrate value
Test first. Then decide on adoption.
Before a pilot, we agree baselines and success criteria. These may include repeat checks avoided, time to the next agreed step, decision quality and correct application of learned rules in later tasks.
Knowledge capture, integration, review, workload and running costs also count. Time released is initially capacity, not an automatic cash saving. Quality and safety requirements remain prerequisites.
Measured value − Setup and operating effort
Planned pilot boundaries
Traceable support. Human responsibility.
What “Verified” means
Task-specific checks of sources, versions, scope and evidence. This is not general certification or a guarantee that every error will be detected.
Advisory in operations
EVA should flag uncertainty and involve the responsible people. In the planned pilot it controls neither equipment nor workflows and does not authorise production.
Data & knowledge
Controlled access, with no cross-company sharing by default. Existing intellectual property remains with its owner; rights to new results are agreed.
Designed with employees
Informed participation, minimum data and no individual performance monitoring. Interaction, workload, accessibility and potential bias are part of the assessment.
Technical starting point
Start with existing knowledge.
Approved instructions, videos and expert conversations can provide the foundation. Structured asset context through the AAS and operating data through interfaces such as OPC UA or MQTT can be added where the task benefits.
On-site operation, European hosting and suitable models would be assessed with the industrial partner. Model origin and hosting location are separate decisions. Voice, text, images or video should suit the work environment; no avatar is required.
View the EVA overview (German labels)
An invitation to industrial partners
45 minutes to explore the fit.
Someone responsible for production, maintenance or service brings a short description of a recurring problem. We discuss who is involved and how you would recognise an improvement.
Explore the fit
No preparation, files, NDA or budget commitment required. Keep confidential details out of this first conversation.
Define a bounded task
If both sides are interested, process experts and IT/OT help clarify the task, data, interfaces and effort.
Agree and assess
Before committing to a pilot, agree resources, costs, rights and stop criteria. The aim is a prototype, assessed evidence and a basis for a decision.
The SMC form records non-binding interests and enquiries, not appointment bookings. Newsletter consent is separate and optional. The form is in German; you can also email Peter in English.
Project status · October 2026
Prepare the research together.
EVA targets HORIZON-CL4-2027-02-DIGITAL-EMERGING-52-two-stage. The proposal and partner contributions are in preparation. A funded pilot depends on selection, the grant agreement and partner agreements.
SMC’s proposed role is practical knowledge capture and adoption. Any independently commissioned implementation would require a separate scope and funding agreement.
Background in English
Industry presentation · PDF · 13 pages ↗Background document · PDF · 10 pages ↗More on the pilot, research approach and illustrative value calculations. The model calculations are not savings forecasts.