Stripes, supported by the Public Administration Intelligence (PAI) project

A RAG system to make the most of Stripes' document archive

Gnucoop built for the social cooperative Stripes a RAG (Retrieval-Augmented Generation) system that makes the organization's document heritage searchable in natural language: the projects submitted to tenders and the articles of the magazine Pedagogika.

The project developed in three phases. In the first, documents were indexed and enriched with structured metadata (title, place, type, year) extracted automatically by a language model, building an organized document database. In the second phase, texts were vectorized and split into semantically coherent chunks, making them searchable by meaning as well as by keywords; the same logic was applied separately to the Pedagogika article archive.

The project ended with a search interface that allows filtering documents by metadata and running semantic searches on the full text. The answers generated by the system always cite the source documents they come from, guaranteeing traceability and verifiability of the results.