← Cases
United StatesEducation2025 – ongoing

Carnegie Mellon University

ScottAI library reference and discovery agent

Carnegie Mellon's ScottAI logged 1,414 first-year sessions and gained live library connections after giving outdated answers

In progressCompany's own claimOfficial ↗

Problem

Library users needed help finding resources without knowing which database to search. An earlier rule-based chatbot restricted questions to predetermined menus.

Approach

The Libraries built ScottAI internally and launched it in July 2025. After incorrect answers exposed weaknesses in its website-based information, staff connected it to live library systems through Google Vertex AI Agent Builder. Librarians tested responses and refined its search instructions.

Results

1,414sessions in ScottAI's first year, reported by the universityCompany's own claim[1]

Takeaway

Access to authoritative material did not prevent outdated answers. Maintaining current connections and testing reference questions became part of development.

Categories

Industry
Education
Business function
Customer service
Innovation type
Process innovation
Evidence
Company's own claim

Sources

Source summaries

[1]Carnegie Mellon University LibrariesOct 2026

The Libraries reported ScottAI's launch and first-year usage. It described incorrect answers, subsequent connections to live systems and librarian testing.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026
Open case page ↗