About GLADYS

GLADYS stands for GenAI Library Assistant, Directing Your Search. It was developed by Librarians at SDSU as a research collaborator, not a replacement for expertise. It can help explore a topic, identify resources, and develop research strategies.

AI-generated responses can be incomplete or incorrect. GLADYS encourages you to critically evaluate results by providing links to source materials. For additional help, you can always Ask a Librarian.

How GLADYS Works

flowchart LR
    accTitle: GLADYS workflow
    accDescr: A question is checked for a known item and spelling to create a search string with facets for Primo. Topic and request type determine the presentation template. Librarian, database, research guide, and FAQ knowledgebases supply AI-selected recommendations. All results feed the GLADYS Dashboard.
    Q(["💬 Ask a question"]) --> TITLE["Known item identified"]
    Q --> SPELL["AI checks spelling"]
    Q --> TOPIC["AI identifies the topic"]
    Q --> TYPE["AI identifies the request type"]

    TITLE -->|Yes| JOIN["Search String Created"]
    TITLE -->|No| BOOL["Fine tuned AI model<br/>creates Boolean string"]
    BOOL --> JOIN

    SPELL --> FACETS["Facets Applied"]
    FACETS --> JOIN
    JOIN --> CAT[("📚 Primo Search API")]

    TOPIC --> SPLIT["Presentation template selected"]
    TYPE --> SPLIT

    SPLIT --> LIB[("RAG knowledgebase<br/>Librarians")]
    LIB --> SELECT["AI selects the<br/>best librarian"]

    SPLIT --> DB[("RAG knowledgebase<br/>Databases")]
    DB --> DBSELECT["AI selects the<br/>best databases"]

    SPLIT --> GUIDE[("RAG knowledgebase<br/>Research guides")]
    GUIDE --> GUIDESELECT["AI selects the<br/>best research guides"]

    SPLIT --> FAQ[("RAG knowledgebase<br/>Library FAQs")]
    FAQ --> FAQSELECT["AI selects the<br/>best FAQs"]

    CAT --> RESULTS(["✨ GLADYS Dashboard"])
    SELECT --> RESULTS
    DBSELECT --> RESULTS
    GUIDESELECT --> RESULTS
    FAQSELECT --> RESULTS

    classDef ai fill:#eee6ff,stroke:#8057bf,color:#302044;
    classDef source fill:#e4f1ff,stroke:#4788bd,color:#17364e;
    classDef process fill:#e5f4e9,stroke:#569168,color:#203f29;
    classDef terminal fill:#24364b,stroke:#24364b,color:#ffffff;

    class SPELL,TOPIC,TYPE,TITLE,BOOL,SELECT,DBSELECT,GUIDESELECT,FAQSELECT ai;
    class CAT,LIB,DB,GUIDE,FAQ source;
    class SPLIT,FACETS,JOIN process;
    class Q,RESULTS terminal;

GLADYS creates a Primo search string and selects library recommendations for the dashboard.

GLADYS is built on Microsoft Azure-hosted AI services, where it uses gpt-4o and a fine-tuned Llama-3.3-70B model to deploy steps. Retrieval is handled through Azure AI Search, and pulls from several SDSU Library systems:

OneSearch records for books, articles, media, newspapers, and other library materials.
Springshare / LibGuides data for research guides, subject metadata, databases, and Librarians.
LibAnswers FAQ data for library service questions.

When a user submits a question, GLADYS classifies the query by subject area and request type. It also checks whether the query looks like a known item, such as a specific book, article, journal, or citation.

For library material searches, GLADYS generates a Boolean search strategy, applies appropriate facet limiters when detected, and searches OneSearch. For librarians, databases, guides, and FAQs, GLADYS normalizes questions into search-friendly terms, searches relevant indexes, then enriches results with Springshare data where available.

Accessible and Inclusive

GLADYS is tested for WCAG compliance and built on Bootstrap 5 to reduce barriers to use.

Users can report inaccurate, biased, harmful, or otherwise problematic responses using the feedback feature.

Responsible and Sustainable AI

GLADYS is designed to use AI efficiently by considering cost and computational impact. It is designed to use the minimum tokens required.

Try GLADYS Ask a Librarian Use Non-AI Search

Your Privacy

Using GLADYS is optional. You can access SDSU Library resources without generative AI through OneSearch, Databases, Research Guides, and Ask a Librarian.

GLADYS is designed to minimize the collection and retention of personal information. Please do not enter confidential, sensitive, or personally identifiable information. Questions may be logged and used for tool development. Information will be used in the aggregate for research purposes.

Azure OpenAI states that prompts, completions, embeddings, training data, and fine-tuned models remain exclusive to the customer. They are not shared with other customers or model providers, and are not used to improve Microsoft, third-party, or foundation models without explicit permission or instruction. Learn more about how Azure OpenAI processes data.