LUMEN D3.4 Co-design of LLM/AI-Human Interaction

This deliverable (D3.4) reports on the results of co-design user research for LUMEN, in particular the design of the LLM-Chatbot services. The project specific objective relevant to WP3 is the following:

SO2 - Enhance the qualitative discovery of research outputs by establishing an ecosystem of innovative features. 

Following initial one-to-one interview research with potential users, previously developed and accepted design artefacts, user personas and user scenarios as reported in D3.1 (De Paoli et al., 2025), were used as empirical grounding for co-design workshops. Rather than introducing new assumptions, these artefacts acted as placeholders and prompted discussion to assist co-designers to understand ways to design functionalities for the white label discovery platform (WLDP) - LUMEN. Firmly positioned in users’ needs and practices previously identified, each co-design workshop was crafted to explore the design of one functionality at a time. While each workshop centred on one specific functionality, the analysis moved beyond a feature‑by‑feature perspective by synthesising findings across workshops, foregrounding shared user needs, cross‑functional dependencies, and system‑level considerations. 

In this report we include the co-design workshop material relevant to the design of the WLDP LLM-driven services. In total, across D3.3 and D3.4, Twelve co-design workshops were held to understand service needs with potential users, adhering to FAIR principles. This report highlights three of those workshops, which focused on the design of Large Language Model (LLM) services tailored for four core scientific domains: Social Sciences and Humanities (SSH), Mathematics, Earth Systems (ES), and Molecular Dynamics (MD). The online workshops acted as a shared space allowing individuals and domain-level communities to explore and unravel meaning for user-designed system approaches.  This work is vital for ensuring adoption and engagement in a novel system which has expanded target users groups and objectives since initial TRIPLE research with a SSH focus (Dumouchel et al., 2020). 

Methodologically, this iterative method, which builds on previous work,  is highlighted in respected HCI work (Dix, 2009), where understanding user requirements systematically informs subsequent stages of design and implementation.  At its core, this approach reflects the software design lifecycle, where the systematic analysis of user requirements informs and guides subsequent stages of design and development. The rigour of this research is further demonstrated through the provision of actionable insights that support implementation by GUI designers and developers provided in Section 5 onwards. In summary, psychological and socio-technical grounding of the work in this report contributes to UI prototyping (shown in Section 6), and the implementation of this, will predict the usability and adoption of the WLDP.  

Furthermore, guided by FAIR principles, LUMEN’s user-centred design approach promotes democratisation, autonomy, and user ownership by ensuring the system designs reflect users' contextual needs and practices across diverse scientific communities. As LUMEN will incorporate a wide variety of resources, it has been essential to understand how researchers will interact with these tools, extensibilities, semantics, and governing principles as the system is being developed. The term coined for improving validation is the Formality Gap (Dix, 2009), which captures the processes of understanding formal systems against real human needs and practices. Addressing this gap requires attention and the invitation to listen to subjective experiences to reduce the distance between perceived needs and machine-level processes. In our co-design workshops, we facilitated, designed, and hosted a space to invite researchers to share these needs.

This deliverable is currently under review by the EC.