Banca de QUALIFICAÇÃO: THAÍS DE ARAÚJO DE MEDEIROS

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : THAÍS DE ARAÚJO DE MEDEIROS
DATE: 17/02/2025
TIME: 08:00
LOCAL: Remoto
TITLE:

A Language Agent-Oriented Methodology for Automotive Assistance: Integrating Prompt Engineering and MLOps in Advanced Chatbots

KEY WORDS:

Retrieval-Augmented Generation, Automotive Assistance, Prompt Engineering, Language Models, Self-RAG, Gradient Descent.

PAGES: 80
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUBÁREA: Eletrônica Industrial, Sistemas e Controles Eletrônicos
SPECIALTY: Automação Eletrônica de Processos Elétricos e Industriais
SUMMARY:

Technological advancements have significantly transformed the way technical knowledge is sought and interpreted, especially in complex domains such as automotive assistance. In this context, language models employing Retrieval-Augmented Generation (RAG) techniques stand out for enabling the retrieval and generation of accurate responses from large volumes of textual data, such as automotive manuals. However, the increasing complexity of search scenarios demands more robust solutions capable of integrating continuous optimization and dynamic adaptation of queries and responses. In light of this, this study proposes a methodology oriented toward language agents for automotive assistance, combining Prompt Engineering and MLOps practices in advanced chatbots. The approach comprises four variants: standard RAG, RAG with Gradient Descent (for prompt optimization), Adaptive Self-RAG, and Adaptive Self-RAG with Gradient Descent. The results show that standard RAG produced more accurate and structured responses, while RAG with Gradient Descent exhibited shortcomings in precision and scope. Adaptive Self-RAG and its variation demonstrated potential but faced challenges in clarity and precision, requiring refinements to balance self-criticism and content generation.

COMMITTEE MEMBERS:
Presidente - 2885532 - IVANOVITCH MEDEIROS DANTAS DA SILVA
Externo ao Programa - 2249146 - CARLOS MANUEL DIAS VIEGAS - UFRNExterno ao Programa - 2353000 - ELIAS JACOB DE MENEZES NETO - UFRNExterna à Instituição - MARIANNE BATISTA DINIZ DA SILVA - UFAL