What is the (potential) role of generative AI in RSE training & research?

What is the (potential) role of generative AI in RSE training & research? Opportunities, Challenges, and Outlook Carlos Utrilla Guerrero AI Researcher UPM, (former Trainer at TU Delft Univ.) Event: LA Referencia Webinar Date: Nov 20, 2025 [email protected] https://carlosug.github.io/ TU Delft & Research Software Engineering (RSE) -Teaching core RSE skills & activities in courses & workshops for researchers -Developing intermediate & advanced RSE training courses -Defining GenAI essentials & good practices for RSE TU Delft Digital Competence Centre is helping shape the university´s strategy towards AI and RSE through: Research Data and Software Unit - Training for Researchers Landscape Research Software (RS), Sustainability, and Generative AI (GenAI) RS define as “any software created during the research process or a research propose” [1] Katz [2] viewpoints of RS sustainability “as the process of developing and maintaining software that meets its purpose over time” Uni or bi-directional Scientific Publication Refers to “Any software artefact which is linked to a scientific publication” [3] Software Best Practices & Training Reproducible Endurance Evolvability RS Sustainability How GenAI tools* support Sustainable RS? *End user tool [...] whose technical implementation includes generative based model such as transformer architecture or deep learning [4] (Sorry for the simplification) Can AI* rid us of all research software reuse issues? Comprehend Interpret Integrate Execute Reuse I want to reuse repo: git/nlp Success Conceptual Workflow overview RSE Assistant [5]: carlosug/agent.rse at ete-version Research Goal: Developing AI-based assistant to autonomously transform research software documentation into machine-actionable workflows, reducing researcher burden and accelerating discovery [6] *LLM-powered assistants are relevant in our approach, but interest in formal reasoning (i.e. symbolic methods) is rising again. Neuro-symbolic applications (i.e. merging LLMs with Logical-based reasoning) is increasingly relevant README .md PhD training needs a reboot in an AI world As machines get better at data analysis and software development, doctoral training must evolve DOI: https://www.nature.com/articles/d41586-025-03572-w (Nov 3, 2025) Workshop on exploring GenAI for RS in a nutshell Provide a platform for knowledge sharing & discussion on relevant tools & services for improving RS quality GenAI for RS Overview What is GenAI? What challenges are in RSE? What tools exists & Demo 50 Min | 20 TU Delft Researchers Published material: https://zenodo.org/records/11492084 https://www.tudelft.nl/2024/delft-ai/sp ring-symposium-2024-showcases-aieducation Build RS faster using GenAI tools, but think twice before integrate into the research workflow Purpose: Dependency solver (+) Integrated into IDE (-) Hard verification (-) Limited in harder tasks GitHub Copilot Purpose: General Questions (+) Easy to use (-) No replicability (-) Personal data risk ChatGPT Purpose: System Development (+) access local LLMs (-) fragmented tools (-) lack of standards Gen AI Devs Stack* Purpose: Research Level (+) access to open LLMs (-) Token Limits (-) Limited free version LLMs local Learners are already making extensive use of the tools in their work * https://www.xenonstack.com/hs-fs/hubfs/generative-ai-development-stack.png?width=1280&height=922&name=generative-ai-development-stack.png Towards responsible adoption of GenAI for RSE Maintaining solid RSE practices while leveraging GenAI tools Understand how to efficiently: 1. Apply prompting strategies 2. Audit models responses ensuring legal & logical & quality standards By openly discussing our challenges, we strengthen the entire RSE community Workshop Survey Results (N = 15) : https://zenodo.org/records/11492084/files/AIforRS-mentimeter.pdf?download=1 Look forward Opportunities for alignment of RSE practices/training initiatives within RDA and beyond Evaluating GenAI models output for RSE - Developing standards for auditing GenAI response - Examining legal & ethical issues for effective adoption Examining suitability of tools and services - Cataloging tools and services (EVERSE Technology Radar) - Experimenting existing technical research stack Exploring new educational approaches - Determining appropriate level of automation in RSE - Reviewing curricula activities to be taught in GenAI for RSE https://everse.software/TechRadar/ https://everse.software/RSQKit/ Reproducibility IG PRO4 RS WG Software Code IG FAIR4 RS WG Gen AI4 RS