Primo Research Assistant is a generative Artificial Intelligence-powered tool integrated into Primo VE, designed to accelerate and simplify academic research. By allowing users to pose natural language questions, the Assistant scopes their institutional library’s holdings to pinpoint up to five highly relevant scholarly articles and creates an overview that includes inline source references. Users can further explore full records and additional sources for deeper verification. All Primo VE customers have access to the tool, though institutions may face volume limitations over time to manage costs.
The system operates using a Retrieval Augmented Generation (RAG) architecture that fuses the expansive linguistic capabilities of a Large Language Model (currently OpenAI’s GPT-4o mini) with the indexed content from the Central Discovery Index (CDI). When a user asks a question, it is first converted into Boolean queries in both the local language and English, then results are retrieved from CDI, re-ranked using embeddings, and distilled into a comprehensive overview. Primo Research Assistant supports multiple languages, though support levels can vary depending on language and content availability, and the summary overview is provided in the language of the query.
The tool’s scope covers all CDI metadata and abstracts except for news content, certain restricted collections, withdrawn/retracted materials, and content from specified major providers such as APA, Elsevier, and JSTOR, pending ongoing stakeholder discussions. Primo Research Assistant is activated at the view level, with configuration options for icon or widget access and can be further refined through detailed settings and labels for localization and accessibility. Users can access the tool via the main menu or a widget on the results page, and must be signed in unless an institution specifically requests broader access. Functionality includes advanced querying, easy application of refinements (type, date, availability), and session-based research history that can be saved and organized by topic. Feedback is encouraged via UI controls and email, and user activity is tracked through dedicated analytics dimensions.
Primo Research Assistant delivers answers with clear references and supports refined searches by resource type, date, and online availability. It preserves user research history (up to 200 searches), enables deletion and organization of topics, and provides configuration settings for privacy, session management, and UI customization. Users are cautioned that, as with all Large Language Models, responses may contain inaccuracies and should be corroborated with original sources. The product is in its beta phase, with ongoing developments based on community input and usage analytics.
