Prompt engineering for library agents: development of an AI-based conversational reading recommendation system

Abstract

This article presents the design, development and implementation of LaIABot, an artificial intelligence agent based on structured prompting for bibliographic recommendations in library environments. The system addresses common limitations such as staff shortages and time constraints, improving the accessibility and personalization of the service. A key innovation is the design of prompts that generate two simultaneous outputs: one for the user, with clear recommendations and another for the institution, in JSON format with metadata from the interaction, which allows for feedback and optimization of the system, integration into SIGB and enrichment of document analysis and management processes. A user-centered approach grounded in library science and ethical principles was adopted. LaIABot is developed with a configurable base prompt composed of five modules: user profile, suggestion filtering, collection typology, library services and feedback mechanisms. The technical architecture employs Flask, HTML5, JavaScript and CSS3, with SQLite for initial validation and the possibility of migration to more robust database managers. The pilot test, using 150 synthetic queries and a corpus of approximately 8,500 records, evaluated precision, diversity and JSON output consistency, achieving a 99.33% structural success rate and a 96.67% effective retrieval rate. Libraries are encouraged to explore the use of specialized conversational agents as a reader advisory tool, prioritizing structured prompting tailored to their specific context and collections. Documenting and auditing recommendations is essential to ensure authorial and thematic diversity, alongside implementing ethical oversight and data governance protocols. The dual output (user/JSON for institutional use) should be leveraged for usage analysis and catalog enrichment. The model's open-source and modular architecture supports its replication in public, academic and specialized libraries, enabling progressive AI integration aligned with institutional infrastructure and resources. LaIABot extends library service availability by operating 24/7, personalizes recommendations according to user profiles and integrates with existing systems, supporting reader advisory without requiring significant investment. Its JSON dual output enables integration with ILS platforms, statistical analysis and continuous service improvement. Pilot test results indicate adequate average response times, thematic and temporal diversity of authors and robust data structuring, confirming its viability as an applicable prototype. The system can enhance the visibility of historical collections and lesser-known authors, strengthening the library's social and educational role. This work introduces a pioneering approach to applying structured prompting with dual output for bibliographic recommendation. Unlike generic AI solutions, it offers a parameterizable architecture grounded in library science and ethical standards, capable of simultaneously producing valuable user-facing responses and structured institutional data. Its open and replicable nature, combined with technical validation in a controlled environment, lays the foundation for deployment in diverse contexts. This research advances the integration of generative AI into library services, combining technological innovation with the preservation of professional values such as neutrality, inclusion and equitable access to knowledge.