Veolia Italia partnered with auticon to build an LLM-based agent that turns unstructured PDF documents into structured, analysis-ready data.
Veolia’s Italian operators had to manually search for data inside multi-POD energy bills, rebuild tables in Google Sheets, and handle different formats for each supplier. Bills could run to thousands of PDF pages with heterogeneous, non-standardised structures. The goal was to make this data quickly accessible and analysable to support energy audits, technical reports and regulatory reporting in a simple, scalable and replicable way.
auticon developed an LLM-based agent that transforms digital PDF documents into structured data. The user uploads one or more documents, defines the fields to extract, and the agent automatically generates an analysis-ready table. The application was built in Python with LangChain, integrated with Streamlit for rapid front-end development, and embedded in a scalable framework for deploying Generative AI applications. Initially built for energy audits, it quickly proved effective on multi-supplier bills, invoices, technical reports and cybersecurity documentation.
Showcased how a neurodivergent auticon consultant brought a structured, modular working method that became a best practice within Veolia, while delivering tangible business value through Generative AI.
