AgroNLP – Web application for natural language queries (NLP) on agricultural databases using the Deep Seek api

Abstract

The development of artificial geointelligence represents a technological advance in projects managing large-scale geospatial data. In Cuba, initiatives like the agricultural cartographic database project stand out. This paper introduces AgroNLP, a web application enabling natural language queries on a PostgreSQL/PostGIS agricultural database, generating NL responses via the DeepSeek API (deepseek-r1:free). The solution integrates an LLM fine-tuned to interpret queries and generate contextualized SQL, following an MCP-like approach. Built with Next.js, PostgreSQL/PostGIS, and Deep Seek, it removes the need for SQL expertise. Its novelty lies in agricultural specialization and geospatial capabilities, outperforming generic tools like Claude [4].