Ask a robot - adding AI search to a repository

Abstract

This presentation demonstrates practical implementation of AI-powered search in academic repositories, moving beyond the "put an AI in it" hype to examine real capabilities and limitations. Using public data from Project Gutenberg, I'll showcase how Large Language Models (LLMs) enable semantic search that surpasses traditional keyword matching and manual tagging by understanding thematic relationships between texts.

The session includes live demonstrations (pre-recorded) comparing abstract-based versus full-content similarity search, and entertaining examples of AI agents exceeding their intended scope when faced with inappropriate queries. These concrete examples from a working system illustrate both the promise and pitfalls of AI integration.

As repository developers increasingly receive requests for AI-enhanced features, this talk provides practical context for evaluating these tools. I'll share a documented, reproducible path for experimentation, addressing both technical implementation and broader concerns including environmental impact and the democratisation of AI development—critical considerations for Open Science in an increasingly closed technological landscape.

The presentation concludes by examining how AI chatbots are reshaping human-computer interaction and what this means for maintaining accessible, user-centered scientific information systems in an era of changing user expectations. Disclaimer: my A.I. wrote this.