CERTAIN Pilots

CERTAIN Project: Pilots presentation

The CERTAIN project validates its framework for AI compliance through seven pilots across six critical business sectors. Using a Living Lab methodology, these pilots ensure AI solutions are technically proficient, ethically aligned, and fully compliant with the EU AI Act and GDPR.

Each pilot follows a three-step activity cycle: Require (defining needs), Living Lab (research and tool development), and Validate Operationally (real-world testing against KPIs).

Pilot 1

Biometric Systems for Border Control

Biometrics / Security

Developing AI-driven biometric recognition (facial data) for high-risk border management (1:1 passport checks) and airline boarding (1:N passenger checks).

To test a “black box” setup for fairness, robustness against identity fraud (morphing/inference attacks), and the trade-off between energy efficiency and performance.

Fairness, Privacy, and Trust.

Bias Reduction: ≥50% reduction in demographic bias.
Energy Efficiency: Tenfold reduction in energy consumption compared to Q1 2024 solutions.
Security: Maintain error rates for presentation and morphing attacks below 10%.

Pilot 2

AI Support for Childhood Psychiatry

Healthcare

Utilizing explainable AI and multimodal data (video, audio, sensors) to enhance early detection and monitoring of psychiatric risks in youth and creating privacy preserving Data Sets for secondary use.

To validate clinical impact through Living Labs, ensuring the system provides transparent “risk scores” for shared decision-making while preventing demographic bias.

Ethics, Clinical Trust, Autonomy, and Transparency.

Bias & Accuracy: Reduce diagnostic bias by 30% and achieve ≥70% sensitivity and specificity.

Acceptance: ≥85% clinician trust and ≥70% trust from parents/children.

Efficiency: 30% reduction in time to diagnosis [Pilot 2 Measurable Improvements].

Pilot 3

AI-Driven Energy Planning

Energy / Sustainability

AI systems for Renewable Energy Communities (RECs) to manage demand forecasting and consumption optimization for households and small businesses.

Evaluate GDPR compliance regarding personal energy data and assess AI fairness to ensure optimization models benefit small producers, not just large ones.

Accountability, Inclusion, Trust and Reliability.

Efficiency: 80% reduction of energy surplus [Pilot 3 Expected Improvements].

Adoption: 30% increase in trust and adoption of automation solutions [Pilot 3 Expected Improvements].

Pilot 4

EU Human Capital Market Demand Analysis

Human Resources (HR)

Enabling the trusted and secure publishing of sensitive HR data (GDPR, EU Legal Requirements) among the multiple actors in the HR domain within the Dataspace. Showcasing the added value of Large Language Models (LLMs) through a chat based interface to automate candidate matching, improve recruitment efficiency and labor market trends forecasting.

To test the readiness, usability and compliance of the CERTAIN data space for data holders/end-users, focusing on automated regulatory compliance and data sovereignty.

Data Sovereignty and Explainability; Interoperability & Trustworthiness; Security & Privacy; Transparency and Usability.

Matching: 30–40% increasing the matching of professionals with job opportunities

Skills: 20-30% Enhancing skills and awareness in high-demand sectors.

Forecasting: 20–30% improving the ability to forecast Human Capital trends in Labour market and education sectors [Pilot 4 Expected Improvements].

Pilot 5

Compliance Guidance for Data Holders

Data Economy / SMEs

Supporting SMEs in navigating legal and ethical complexities when entering AI data markets through tailored compliance “wizards”.

To evaluate the usefulness and accessibility of CERTAIN’s simplified guidelines for non-expert users to reduce the burden of EU regulatory compliance.

Accessibility, Demonstrating Compliance, and Cost-Effectiveness.

Usability: System Usability Scale (SUS) score > 68.

Efficiency: Significant reduction in time and cost of compliance checks compared to manual audits.

Pilot 6

Personalized Investment Recommendations

Finance / Retail Banking

Generating personalized (1to1) investment recommendations aligned with customers’ risk profile.

To develop and test counterfactual explanation methods (identifying how different inputs change decisions) to reveal algorithmic biases and improve trustworthiness [Pilot 6 Evaluation Purpose].

Suitability, Transparency and Explainability.

Reliability: Achieve at least 80% accuracy and suitable recommendations.

Speed: Recommendations generated within 20 seconds.

Pilot 7

Automating MLOps for regulatory compliance in digital solutions deployment

Information Technology (IT) / RegOps , MLOps

Delivering AI-boosted and compliant-by-design technical solutions and products that adhere to EU regulations.

To evaluate the effectiveness of streamlining the compliance process in addressing legal, ethical, and operational challenges in a product deployment environment.

Compliance, Efficiency, Ethics, Risk Reduction, and Market Advantage.

Compliance: ≥90% of digital solutions successfully pass compliance checks.

Cost & Speed: ≥15% reduction in compliance costs and ≥60% reduction in time for security assessments.