Green Innovation Extraction
Published in GRINS – Growing Resilient, Inclusive and Sustainable, Spoke “Innovation: Ecosystems for Circular Economies”; in collaboration with University of Verona, 2026
A pipeline for extracting and classifying “green innovation” claims and green product labeling from corporate annual reports, scanning and analyzing 40+ Italian listed companies’ annual reports. Combines automated environmental-relevance filtering (BERT-based scoring) with LLM-driven extraction and multi-label classification against established green/circular-economy taxonomies, with full database-backed traceability from source text to finding.
In collaboration with the University of Verona (Cristina Vlorio et al.), funded by GRINS – Growing Resilient, Inclusive and Sustainable, Spoke “Innovation: Ecosystems for Circular Economies.”
Technology: Python pipeline with SQLAlchemy/Alembic-backed persistence and a Chroma vector store, combining transformer-based environmental-relevance scoring (EnvironmentalBERT/ClimateBERT) with LLM-driven extraction (OpenAI/Anthropic APIs) and multi-label classification of corporate disclosures, built with a domain-driven, onion-architecture design.
My role: Tech lead for GRINS; wrote the whole pipeline end to end: the full extraction, classification, and traceability process.