Maximizing Productivity at the Midway Project Meeting in Copenhagen

This week THEMIS project partners gathered in Copenhagen at the Democracy-X offices for a two-day project meeting, an important milestone marking the midway point (18 months) into our mission to create a system that helps professionals assess and improve the trustworthiness of AI.

Team THEMIS outside the Democracy-X offices in Copenhagen.

Tailoring Trust: Trustworthiness Isn’t One-Size-Fits-All

At the core of THEMIS is the belief that trust in AI depends on context. A healthcare professional may prioritize accuracy above all else, while a legal expert might be more concerned with fairness. Because optimizing for every characteristic at once can lead to trade-offs, THEMIS allows users to focus on the dimensions of trustworthiness that matter most in their domain.

Through an interactive chatbot, users are guided to articulate their specific needs, which then shape a tailored risk assessment. THEMIS includes eight components in total, from risk modelling to recommendations—each playing a role in this modular decision-support system.

A Mirror for AI: THEMIS as a Socio-Technical Replica

THEMIS works by replicating the socio-technical environment of the system under test (SUT), giving users insights into potential risks and areas for improvement. Much of the meeting focused on maturing the components and preparing for the next major release. Since each partner currently hosts their own data and software, we discussed how to deliver a unified user experience without a shared infrastructure.

A proposed solution could be a federated web portal with a single access point for login, user management, and seamless navigation. Using tools like Keycloak and Webpack, we could implement federated identity management for secure, token-based single sign-on across all components. Micro-frontend architecture will ensure a consistent look and feel, while each partner retains control over their backend and APIs.

Challenges remain, particularly around differing technical stacks across partners (e.g. Spyderisk supports Keycloak, while others rely on REST/JSON), but there was strong momentum to resolve these soon.

Pilots & User Engagement: Deepening Dialogue

We also advanced plans for piloting, with two workshops per use case to deepen engagement with real users. The first workshop will provide initial feedback on components that make up the THEMIS solution. For example, in the healthcare use case, users will interact with THEMIS as applied to an AI system designed to help diagnose pancreatic cancer. Discussions will explore how the components support decision-making, how outputs should be presented, and where improvements are needed. Feedback will be cascaded back to technical teams for refinement. Democracy-X’s flexible platform will support this evaluation process, with the ability to localize and translate content. Baseline questions will be used to assess both individual components and the overall system.

Looking Ahead After Copenhagen

The THEMIS project is entering an exciting new phase. With foundational components in place and early user feedback on the horizon, we’re well-positioned to shape a practical, transparent, and adaptable framework for AI trustworthiness—tailored to the real needs of professionals across sectors.

Stay tuned for more updates as we move toward our next release and continue piloting across healthcare, ports, and media.