Igor de O da Silva

Igor de O da Silva

Auto-generated profile, based on publicly available information

Igor de O da Silva

Universidade Tecnológica Federal do Paraná, Brazil

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 January 2026 in Revista Brasileira de Engenharia Agrícola E Ambiental
Verified
Journal Article
Open access
GLAUCO Vieira MIRANDAGLAUCO Vieira MIRANDA
GLAUCO Vieira MIRANDA
ISIgor de O da Silva
Igor de O da Silva
ABArlete T Beuren
Arlete T Beuren
Advances in artificial intelligence and natural language processing are transforming decision-support systems in agriculture, enabling real-time interaction between humans and machines for smarter and more accessible crop management. This study presents the development and evaluation of an intelligent monitoring system for hydroponic cultivation, integrating a chatbot based on large language models (LLMs) and a retrieval-augmented generation (RAG) architecture. Designed to support smallholders, the system provides real-time technical assistance in natural language by integrating environmental sensor data, such as pH, temperature, and electrical conductivity (EC), with a semantic vector database indexed by four embedding models: text-embedding-ada-002, multilingual-v2, all-MiniLM-L6-v2, and Word2Vec. Twenty structured and unstructured agronomic questions were used to assess the system’s ability to generate relevant and factually accurate responses through ChatGPT and Gemini. Real-time data from lettuce cultivation trials were formatted in JSON, which were used to simulate decision scenarios. These included both historical queries (e.g., pH trends) and conceptual questions (e.g., ideal EC range). Results showed that the text-embedding-ada-002 model consistently achieved the highest semantic relevance (mean = 0.88) and factual faithfulness (mean = 0.72 with Gemini), outperforming the other embeddings. The multilingual-v2 showed moderate results, while the MiniLM and Word2Vec performed poorly in complex or context-dependent tasks. Gemini slightly outperformed ChatGPT in producing faithful responses. The proposed system is effective as a low-cost decision-support tool for hydroponics and can be scaled to other domains, such as irrigation and greenhouse management. This study highlights the potential of embedding enhanced LLMs and RAG architectures to democratize access to precision agriculture technologies.