Knowledge Graphs as Infrastructure Enabling Generative AI Integration Across Research Software

As AI tools powered by large language models (LLMs) gain traction in research, one major challenge for RSEs is meaningful integration: How can we connect LLMs to the research software stack—scripts, workflows, datasets, and documentation—to make them context-aware, trustworthy, and useful in real scientific environments? This talk proposes treating knowledge graphs as infrastructure to orchestrate research software ecosystems for generative AI integration. By representing tools, workflows, domain concepts, and research artifacts as a semantic layer, RSEs can unlock new capabilities: Enable retrieval-augmented generation (RAG) grounded in scientific domain knowledge Extend AI agents with on-demand invocation of research software components and workflow logic Improve documentation, reproducibility, and transparency in AI-driven research processes The ideas presented are informed by my 15 years of experience in developing data and AI solutions for industrial R&D. I will walk through the architecture of such a system and share implementation examples. In particular, we’ll explore a scenario where tasks are automated by an AI agent operating on top of a knowledge graph and chemoinformatics tools. I’ll also highlight the collaborative role of RSEs and domain researchers in building and sustaining these systems.