Riccardo Dattena
A configurable framework for multi-dimensional evaluation of Generative AI systems.
Rel. Guido Marchetto, Alessio Sacco. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Informatica (Computer Engineering), 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (3MB) | Preview |
Abstract
Generative AI systems are increasingly complex, non-deterministic and heterogeneous, making their evaluation difficult to reduce to a single quality metric or assessment based only on the final answer. As they spread into real-world applications in many different domains, the need for a comprehensive observability and evaluation framework grows. Existing approaches often focus on individual dimensions or final response quality, while real AI systems require evaluation across multiple dimensions and analysis of their behavior and execution. Retrieval quality, consistency, task effectiveness, fairness are only some of the many nuances of AI evaluation. This thesis presents a modular and configurable evaluation framework for heterogeneous Generative AI systems.
The framework is designed to operate in both white-box and black-box scenarios, adapting to the level of system observability available and offering a concrete solution applicable to real-world industrial use cases
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
