Improving core facility service discovery with an AI assistant grounded in institutional web content

Core facilities source advanced technologies and expertise but can remain under-utilized because researchers, students and early-career scientists, struggle to identify relevant units or cannot articulate appropriate technical inquiries. To enhance access to users, we implemented a domain grounded conversational application based on retrieval augmented generation (RAG). It combines advanced general AI-chat behavior with restricted alignment to core-facility services by uniting a Google Gemini File Search as a structured knowledge base and a Perplexity powered web agent for conceptual scientific queries. The system workflow constrains the agents to institutional domains and routes specific operational questions to a file search knowledge base. As a result, it relies on existing core facility websites and is updated in line with them. This article describes the design of the workflow, how the system is optimized including necessary guardrails to prevent general purpose chat. It proposes evaluation metrics such as veracity, cost per interaction, latency, and examples of usage. The chatbot can help researchers better define their experimental needs, discover relevant facilities they had not previously considered, and thereby increase the visibility and accessibility of institutional research infrastructures.