Institutional Repository of Vilnius University
Didžiųjų kalbos modelių grįstų agentinių sistemų projektavimo sprendimų ir reikalavimų įtaka jų realizacijai ir naudotojo sąveikai /
Abstract
dc:descriptionThis work examines the design of a large language model (LLM) based agentic system through a case study. The object of investigation is a self-developed prototype -- an automotive parts inquiry system with an LLM agent interface integrated with the external TecDoc catalogue. First, following ISO/IEC/IEEE~29148 with MoSCoW prioritisation, 8~functional and 8~non-functional requirements were derived, extended with agent-specific dimensions (tool schemas, reasoning-iteration bounds, dialogue initiative). Then eight key design decisions (PS1--PS8) covering execution, dialogue strategy, tool registry, state synchronisation, context management, error resilience, security, and deployment were identified, documented, and evaluated using a five-criterion analytical framework. The resulting prototype was verified through an eight-layer testing strategy (520 passing tests). A bounded empirical check was also executed: 16 completed sessions were collected (8 humans and 8 LLM agents), with the primary human cohort (n=6) yielding SUS~94.17 and Raw~NASA-TLX~7.58. The work argues that the classical requirements framework is insufficient for agentic systems and that API-first execution combined with a schema-driven tool registry are central design decisions shaping user-interaction quality within the bounds of this single case.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Merkelis, Patrikas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308337990&prefLang=en_US