Desenvolvimento de um assistente agronômico conversacional baseado em LLM
Guardat en:
| Autor principal: | |
|---|---|
| Data de publicació: | 2025 |
| Format: | Bachelor thesis |
| Idioma: | por |
| Font: | Repositório Institucional da UPF |
| Download full: | https://repositorio.upf.br/handle/123456789/9874 |
Sumari: | The agricultural sector faces challenges in managing complex technical knowledge, specifically regarding the operation of the Decision Support System for Agrotechnology Transfer (DSSAT). This work presents the development of a conversational assistant based on Large Language Models (LLM) to provide technical support to DSSAT users. The system utilizes a Retrieval-Augmented Generation (RAG) architecture, orchestrated via low-code tools, to process queries via text, voice, and image through a messaging app. The methodology involved curating a specialized knowledge base and an automated evaluation using the LLM-as-a-Judge paradigm on a set of 30 questions. Results indicate high efficacy, with average scores of 0.88 for Faithfulness and 0.91 for Answer Relevancy. The system demonstrated strong capabilities in troubleshooting and conceptual explanations while maintaining agronomic safety by correctly refusing to answer about undocumented parameters. Conclusions suggest that the proposed architecture effectively democratizes access to specialized agricultural knowledge, serving as a scalable technical support tool that reduces entry barriers for new users. |
Sigues el primer a deixar un comentari!