Brightcon 2026, hackathon and courses

Highlights

• AI-assisted workflow for extracting life cycle inventory data from
  secondary resources
• Use of open-source LLMs and fine-tuned
  classification models for document screening and information
  extraction
• Harmonization of heterogeneous inventories for emerging
  technologies with low technology readiness levels
• Transparent and reproducible workflow implemented in Python using
  open datasets and shared code repositories

Description

Emerging technologies often lack primary inventory data, making Life Cycle Assessment (LCA) particularly challenging during early-stage development. As a result, practitioners frequently rely on scientific literature and technical reports to construct Life Cycle Inventories (LCIs). Although relevant information is available across published studies, extracting and structuring inventory data remains a manual and time-intensive task requiring substantial domain expertise. In addition, inventories reported in literature often differ in system boundaries, assumptions, process conditions, and level of detail, complicating their direct reuse and comparison.

This work presents an AI-assisted workflow for automated extraction and harmonization of LCI data from scientific literature for emerging technologies. The proposed framework combines machine learning techniques and open-source Large Language Models (LLMs) to support literature screening, information extraction, and inventory harmonization. The workflow consists of five stages: (1) automated literature collection using APIs and curated keyword sets; (2) identification of relevant publications using a fine-tuned classification model; (3) domain adaptation through pretraining on selected sustainability and LCA-related texts; (4) extraction of inventory data using fine-tuned LLM-based models; and (5) harmonization of heterogeneous inventories into a consolidated inventory suitable for subsequent LCA modeling.

The workflow is implemented in Python using open-source models and datasets to ensure transparency and reproducibility. Codes, metadata, and workflows are intended to be shared through institutional repositories to support reuse and community collaboration. The presentation discusses methodological challenges related to heterogeneous literature-derived inventories and highlights how AI-assisted workflows can support scalable and reproducible inventory generation for emerging technologies.