Sistema inteligente para o monitoramento do cultivo hidropônico indoor

Abstract

This work presents the development of an intelligent system for monitoring indoor hydroponic cultivation, focusing on lettuce production and aiming to assist producers in controlling critical environmental parameters such as pH, temperature, and electrical conductivity. The motivation lies in the need for accessible solutions that automate crop monitoring, reduce losses, and improve production efficiency. To address this, a chatbot was proposed, integrated with a large language model, LLM capable of interpreting real-time data and interacting with users. The methodology involved data collection through sensors, preprocessing and conversion of these data into a structured format, as well as the use of different embedding models for indexing in a vector database. Combinations of four embeddings including text-embedding-ada-002 and two language models, ChatGPT and Gemini, were evaluated using metrics such as context precision, recall, response relevance, and faithfulness. The results showed that text-embedding-ada-002 achieved the best overall performance, while Word2Vec was the least effective. Both LLMs demonstrated the ability to generate coherent responses, with ChatGPT standing out in open-ended questions. The project demonstrated the potential of integrating Artificial intelligence, AI and precision agriculture. This system represents a promising alternative to make hydroponic monitoring smarter, more accessible, and personalized.