Banca de DEFESA: MATHEUS GOMES DINIZ ANDRADE

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : MATHEUS GOMES DINIZ ANDRADE
DATE: 19/12/2025
TIME: 09:00
LOCAL: Remoto
TITLE:

An Agent-Based Methodology with Autonomous Evaluation for Knowledge Extraction from Technical Texts

KEY WORDS:

Retrieval-Augmented Generation, Large Language Models, Multi-Agent Systems, Industrial Network, Knowledge Retrieval, Technical Documentation.

 

PAGES: 75
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

This work develops a methodology for retrieving and synthesizing knowledge from complex technical documentation using generative AI architectures. The approach is investigated by analyzing and comparing three configurations: (i) a Large Language Model (LLM), (ii) a single-agent Retrieval-Augmented Generation (RAG) model, and (iii) a multi-agent RAG configuration. To validate this methodology, a representative case study was applied using technical texts from the PROFIBUS protocol. Performance is evaluated through quantitative linguistic metrics (ROUGE, METEOR, MATTR, BERTScore) and qualitative assessments (LLM-as-a-Judge). The results indicate that multi-agent orchestration improves contextual accuracy and content structure in processing complex industrial texts, demonstrating how generative AI can support knowledge-driven, human-centric automation in line with Industry 5.0 goals.

 

 

COMMITTEE MEMBERS:
Presidente - 2885532 - IVANOVITCH MEDEIROS DANTAS DA SILVA
Interno - ***.640.764-** - MARIANNE BATISTA DINIZ DA SILVA - UFRN
Interna - ***.640.764-** - MARIANNE BATISTA DINIZ DA SILVA - UFRN
Externo à Instituição - DENNIS BRANDAO - UniBs