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

Area of Science:

• Biomedical research infrastructure
• Scientific core facilities
• Research technology access

Background:

• Core facilities offer advanced technologies but are often underutilized.
• Researchers face challenges identifying relevant facilities and articulating technical needs.

Purpose of the Study:

• To enhance user access to core facility services.
• To improve the utilization of institutional research infrastructures.

Main Methods:

• Implemented a domain-grounded conversational application using retrieval augmented generation (RAG).
• Combined Google Gemini File Search for structured knowledge with a Perplexity web agent for conceptual queries.
• Constrained AI agents to institutional domains and operational questions, relying on existing core facility websites.

Main Results:

• The system successfully integrates AI for core facility service discovery.
• Guardrails were implemented to prevent general-purpose chat and ensure domain-specific responses.
• Proposed evaluation metrics include veracity, cost per interaction, and latency.

Conclusions:

• The AI chatbot aids researchers in defining experimental needs and discovering underutilized facilities.
• Increased visibility and accessibility of institutional research infrastructures are key outcomes.
• This approach optimizes resource allocation and supports scientific advancement.