MATRIQS – AI-supported research assistant for materials science
MATRIQS is an AI-supported research assistant for materials science, with an initial focus on 42CrMo4 steel. The aim of the system is to search scientific literature, technical documents and project-specific data in a structured manner, to identify relevant information and, based on this, to generate clear, source-based answers to questions in materials science.
The focus is on supporting research processes in which information from many different references needs to be collated. This includes, for example, details on material states, heat treatments, cooling conditions, microstructures, mechanical properties and experimental boundary conditions. MATRIQS is designed not only to make this information retrievable, but also to classify it technically and link it together transparently.
At its current stage of development, MATRIQS combines a locally controlled document and literature search with AI-supported scientific synthesis. The generated responses are based on the references retrieved and are accompanied by document- and page-specific citations. This ensures that it remains clear on what data a statement is based. The system specifically distinguishes between different material states, heat treatments, cooling conditions, microstructures and the quality of the available evidence.
Particular emphasis is placed on addressing scientific uncertainty. MATRIQS is designed not only to find relevant text passages, but also to indicate whether a statement is directly substantiated, can only be inferred indirectly, or is uncertain based on the available evidence. This is intended to prevent different material states or experimental conditions from being inappropriately conflated.
In the future, MATRIQS is to be expanded into a modular digital research assistant. Planned enhancements include additional scientific data sources, specialised components for literature searches, experimental data, document generation, and interfaces to laboratory and analysis software. In the long term, the system is intended to support materials science workflows, from literature reviews and experimental preparation through to analysis and documentation.
Scientific quality assurance is a central component of the development. Through mandatory sourcing, expert review, regression testing, controlled further development and human evaluation of critical results, MATRIQS is to be established as a transparent, maintainable and reliable digital research assistant. The long-term goal is a system that does not replace materials science research, but rather provides structured support, accelerates it and makes it clearer.