LLM model for ESG Reporting

Sustainability reporting, though an opportunity for organisations to demonstrate their performance with respect to environmental, social, and governance perspectives, has requirements in the transparency of the record, strategy, and policy-making process. However, preparation by way of collection and compilation of data from the reports sometimes poses a lot of difficulties resulting in discrepancies and anomalies. The contribution of this work is a system that applies AI, and in particular, Natural Language Processing, to extract the data of a sustainability report using unstructured reports. An application based on GPT-3 developed by OpenAI transforms a given report from an unstructured format within PDF documents to a structured format according to the ESRS. The operational framework of the system encompasses the following main steps: reading PDF files, text normalization, data categorization, executing requests concurrently, and final refinements. Efficiency in the system is measured by cosine similarity metrics, which check how well the output of the system correlates with manually extracted data. High values of correlation obtained ensure that this system has huge potential to advance sustainability reporting practices. This is one major initiative in realizing transparent.