'RC' chatbot constrained to Resource Central records, presenters say, to limit AI misinformation

Speakers described RC as a rules-based generative-AI research assistant that answers user queries by drawing only on Resource Central records and returns sources inline and at the end of responses.

"RC also leverages probability and pattern prediction, but it adds a rules based layer to ensure accuracy, HRSA aligned mission, and pediatric specificity," Paul Stengel said, explaining the assistant performs two round trips between systems to check content against rules before delivering answers.

Christine Willis demonstrated how RC can be personalized: an attendee who said they were a school nurse could ask RC to create a tabletop simulation for a Connecticut district and download a generated PDF exercise. Presenters emphasized RC is constrained to Resource Central content so results should cite the partner records used to generate responses.

During Q&A, presenters said the chatbot has been available for a couple of years, is experimental and due for an update later this year. They also stressed that quality control for the underlying records is maintained through librarian review, subject-matter experts including pediatric clinicians, and a final HRSA review before posting; translations are linked rather than produced by Resource Central.

Presenters acknowledged the broader problem that general web searches and generative AI can surface low-quality or sponsored content, and positioned RC's constrained architecture and visible sourcing as mitigations rather than guarantees of perfect accuracy. The demonstration showed inline references and links back to the original institutional records for verification.