Sistema de reconocimiento inteligente de facturas aplicado a la gestión de inventarios para cota, estudio de arquitectura e ingeniería.

Abstract

In the construction sector, automating processes leads to significant savings of time and resources, which is why it has become a priority for achieving better results. This project focuses on implementing a web-based inventory management system for the company COTA, integrating optical character recognition technologies and Artificial Intelligence for the automatic processing of invoices, along with a responsive web interface. This proposal addresses the manual handling of inventory and promotes the use of modern technologies. The implementation used the library provided by Ollama to communicate with its server, sending images accompanied by a specific prompt to extract the desired structure with key invoice data: product name, unit of measure, quantity, and unit price. For greater data security and integrity once extracted, the information is presented to the user, who can review each detail to edit, confirm, or reject it before storage. This implementation was carried out using the open-source model Qwen3-VL:2B, which provides an F1 Score of 85.58% for numerical data extraction using the exact matching method for evaluation, while for text extraction it achieved an accuracy of 94.92% using the Levenshtein Distance calculation. Overall, it offers a 66.35% reduction in time for this activity, as the AI is 2.97 times faster than data extraction.