Requirements for an AI-Based Material Recommendation System : A Case Study of Alleima’s Virtual Material Selector

Sammanfattning

Selecting the right material is a critical and increasingly complex engineering decision, and AI-based tools offer a way to make it faster and more consistent. Alleima has built an early prototype of such a tool, the Virtual Material Selector (VMS), but it had not been analyzed against the requirements of a robust and trustworthy recommendation system. This thesis establishes a structured requirements foundation for the continued development of the VMS. The study is a qualitative, exploratory single case study. Twelve semi-structured interviews with internal Alleima stakeholders across sales, R&D, product management, and technical marketing were analyzed using thematic analysis and combined with a prototype analysis and a theoretical framework drawn from multi-criteria decision analysis, decision support systems, trustworthy recommender systems, requirements engineering for AI, and user experience. The findings were translated into a requirements specification, evaluated against the prototype in a gap analysis, and organized into a development roadmap. The work produced 76 requirements across four categories: functional, non-functional, data-related, and UX-related. Because the requirements describe two versions of the tool, a customer-facing version and a planned internal version, each was given two priorities. The gap analysis found that the prototype fully meets 12 requirements, partially meets 19, and does not meet 45. Trust qualities such as verifiable sources and explanations are the prototype's strongest area, while data is the weakest, with no data requirement fully met. The central finding is that the prototype behaves in a trustworthy way but rests on too little data, and that the data layer, rather than the language model or the interface, is where most of the remaining work lies.