LLM-Powered Data Extraction for Digital Product Passports (TwinMind)

TwinMind uses AI and RAG to automate Digital Product Passport creation by extracting data from PDFs into compliant Asset Administration Shells. Designed for SMEs, it ensures high field accuracy and regulatory compliance without requiring complex PLM/ERP integration, protecting market access and competitive advantage.

Key data

Description

The TwinMind project addresses the compliance challenges SMEs face regarding the EU's Ecodesign for Sustainable Products Regulation (ESPR). It focuses on developing an AI-assisted extraction pipeline that utilizes Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) to automate the creation of Digital Product Passports (DPP). By extracting data from unstructured product PDFs—such as datasheets and technical drawings—the system generates compliant Asset Administration Shell (AAS) outputs.

Unlike existing enterprise platforms, TwinMind offers a "zero-integration" deployment, meaning it does not require pre-existing structured PLM or ERP data. The solution integrates the AAS schema directly into the AI extraction process, allowing the system to handle nested structures, unit conversions, and cross-document reasoning. This Innocheque pilot with Elwitec GmbH serves as a functional proof-of-concept to validate field accuracy and completion rates, providing a credible foundation for future R&D scaling.