A Modular IPA Framework for Semantic Retrieval and Generation in Legal-Administrative Workflows

Abstract

The growth of document collections has made manual analysis inefficient in contexts that require synthesis and traceability. This work proposes a framework for orchestrating pipelines in the context of Intelligent Process Automation (IPA), focused on processing unstructured data. The solution integrates OCR, NLP, machine learning, and LLMs within a modular and general-purpose architecture, with support for open-source models. A case study involving more than 14,000 legal-administrative documents demonstrated that the approach enables extraction, semantic enrichment, and vector search. The results indicate gains in efficiency, organization, and support for automated document generation.