Chat and Collection Discovery Working Group
It brings together AI4LAM members interested in how chat-based AI, tool calling, and AI-mediated search are changing discovery, reference, and access to library, archive, and museum collections.
The group provides a community-led, practice-oriented forum for members to share current projects, compare approaches, identify common challenges, and explore opportunities for shared learning and collaboration. Rather than promoting a single model or technology, the group will consider a range of approaches to connecting AI environments with authoritative collections, services, and expertise.
Exploring new approaches to discovery and access
Libraries, archives, and museums are experimenting with new ways for people to discover and interact with collections through conversational AI, semantic and hybrid search, retrieval-augmented generation, AI-assisted reference, and emerging tool-calling protocols such as the Model Context Protocol (MCP).
These developments create opportunities to make cultural heritage collections and institutional expertise more accessible in emerging AI environments, while also raising important questions about trust, provenance, usability, privacy, accessibility, sustainability, and the role of professional expertise.
The Chat and Collection Discovery Working Group will provide a space for AI4LAM members to examine these developments together. Areas of exploration may include:
- How chat and AI-mediated search are changing user expectations around discovery and reference;
- How AI interfaces can connect users with trusted library, archive, and museum collections and services;
- How tool-calling protocols can expose authoritative institutional services to AI environments;
- How institutions can maintain clear connections between AI-generated responses and underlying records, sources, and expert help;
- How to evaluate relevance, grounding, provenance, trust, and usability in AI-enabled discovery systems;
- How institutions are balancing experimentation with privacy, accessibility, sustainability, and professional values; and
- How libraries, archives, and museums can share approaches and infrastructure without duplicating effort.
The group will be vendor-neutral and open to institutions and practitioners at different stages of experimentation and implementation.
Public Call
The date and time of the initial Call for Participants will be announced via email in Fall 2026.
The schedule for recurring meetings will be determined by the facilitators and participating members in coordination with the AI4LAM Secretariat.
The broad vision of the working group is to create a practical, community-led forum for people working on chat, tool calling, and AI-mediated discovery to learn from one another and identify areas where shared work would be valuable.
The group’s priorities and workplan will be shaped by participating members rather than fixed in advance. Early activities may include:
- Community Building — Bring together AI4LAM members working on or interested in conversational AI, collection discovery, reference, semantic search, and tool calling.
- Project Sharing — Provide opportunities for institutions to demonstrate current projects, prototypes, experiments, and lessons learned.
- Landscape Mapping — Identify emerging approaches to AI-enabled discovery and the common technical, organizational, and user-experience questions institutions are encountering.
- Evaluation and Responsible Implementation — Share approaches for evaluating trust, provenance, relevance, usability, privacy, accessibility, and sustainability.
The group will remain demand-driven and non-prescriptive. Specific activities and deliverables will be determined by members based on shared needs and areas of active work.
Specific outputs will be determined by the working group once it is formed. Depending on member interest, activities and resources could include:
- Regular community calls, demonstrations, or show-and-tell sessions;
- Short case studies documenting member projects and lessons learned;
- A landscape scan of chat, tool-calling, and AI-enabled discovery initiatives across libraries, archives, and museums;
- A shared repository of tools, code, implementation examples, or evaluation methods;
- A shared vocabulary or typology describing emerging approaches to AI-mediated discovery;
- A lightweight checklist for responsible implementation of AI-enabled discovery services;
- Presentations, panels, workshops, or other programming for the broader AI4LAM community; and
- Educational programming or training developed in response to member interest.
The working group may explore a wide range of use cases involving AI and cultural heritage discovery, including:
- Natural Language Collection Search — Enabling users to search collection metadata, finding aids, digitized materials, or full text using conversational or natural-language queries.
- AI-Assisted Discovery — Building interfaces that help users explore collections, refine questions, identify relevant materials, and move between generated responses and authoritative records.
- AI-Assisted Reference — Connecting conversational interfaces with institutional knowledge bases, research guides, discovery systems, or other trusted sources to support reference interactions.
- Tool Calling and AI Assistants — Creating services that allow AI assistants to call library search systems, collection APIs, article discovery tools, or reference services on behalf of users.
- Retrieval-Augmented Generation — Grounding AI responses in institutional collections, metadata, documentation, or other authoritative sources.
- Citation and Provenance — Designing interfaces that make the sources underlying AI-generated responses visible and provide clear pathways back to institutional records and collections.
- Hybrid Discovery — Combining traditional keyword search, semantic retrieval, metadata, and generative AI to support different forms of exploration and research.
- Retrieval-augmented generation (RAG);
- Semantic and hybrid search;
- Vector indexes and embeddings;
- Library, archive, and museum discovery APIs;
- Large language model platforms;
- Tool-calling protocols, including the Model Context Protocol (MCP);
- Knowledge bases and institutional repositories;
- Metadata and authority data;
- Evaluation frameworks for AI-enabled discovery;
- User research and usability testing methods; and
- Approaches for measuring grounding, citation quality, relevance, trust, and user outcomes.
The working group will not be tied to any single vendor, model, protocol, or technical architecture.
