UniVERSO does not maintain its own research records — it is a search and summarization layer built entirely on top of VERSO data. At a high level, a query moves through four stages.
On a regular refresh cycle, UniVERSO ingests researcher profiles and publication/creative-work records from the systems that power VERSO (Ex Libris Alma and Esploro). This includes researcher names, titles, departmental affiliations, keywords, research topics, biographies, and — for each output — titles, abstracts, publication venues, and dates. Where VERSO itself is missing an abstract, UniVERSO will attempt to backfill it from open scholarly databases (OpenAlex, with Semantic Scholar as a fallback) so more of the corpus is searchable.
This is a periodic sync, not a live connection — changes made in VERSO are not reflected in UniVERSO instantly. See Editing Your Profile for how to update your VERSO record itself.
Once ingested, publication text and researcher information are converted into vector embeddings — numerical representations that capture the meaning of the text, not just its exact wording. These embeddings are stored in a vector database (ChromaDB) as two parallel indexes: one for individual publications and creative works, and one for AI-generated researcher summaries (see below). This is what allows UniVERSO to match on concepts and topics rather than requiring the same words used in a researcher’s profile.
When you ask a question, UniVERSO embeds your query the same way and searches both indexes for the closest conceptual matches. The candidate researchers that surface are then reranked using several signals together:
Results are also boosted for factors like publication recency and filtered to remove very low-signal record types, so the researchers and works you see are meant to reflect genuine topical relevance rather than raw keyword frequency.
The ranked researchers and publications are handed to a large language model (LLM), which writes a conversational overview of the results, highlights the most relevant researchers and works, and suggests natural follow-up questions. LLM calls are routed through MindRouter, the University of Idaho’s in-house AI platform (with a general-purpose provider available as a fallback), so query traffic stays within university-managed infrastructure rather than depending solely on an outside vendor. You can ask follow-up questions in the same conversation, and UniVERSO will continue reasoning over the same set of results.
In addition to answering live queries, UniVERSO periodically generates a standing summary for each eligible researcher, built exclusively from that researcher’s own VERSO-linked publications, abstracts, and profile fields (keywords, title, department) — the LLM is not drawing on outside knowledge about a researcher’s reputation or work. Each summary is produced in stages: main research themes, methods/approaches, broader impacts, primary areas of inquiry, a mentorship summary (for researchers who have advised theses or dissertations), and a final synthesis combining all of the above. These per-researcher summaries are what UniVERSO searches when it looks for a topical match to a researcher’s overall body of work, separate from matching individual publications.
Because these summaries are generated automatically from limited profile data, they may not capture every nuance of someone’s research — see Managing Your AI-Generated Summary for how to review, revise, or opt out of yours.
Last updated: August 2026