ESGenerator
Extrapolation of unknown data for inclusion in ESG reports.
Every year, companies include what is known as the “ESG report” in the non-financial information section of their annual reports, if they so choose (large companies are required to do so, and SMEs will be required to do so starting in 2027 in the European Union). ESG stands for “environmental, social, and governance,” meaning it represents the company’s environmental, social, and governance criteria.
This report covers all of the company’s social and environmental responsibility activities: the environmental impact of its operations, the decarbonization measures it is taking, the activities it carries out with various associations and nongovernmental organizations, and so on. There are various standards for preparing this report, such as the Global Reporting Initiative [1] or specific ISO standards, but companies have considerable freedom in how they draft it.
A significant portion of the data included in ESG reports pertains to the impact of a company’s operations: electricity consumption, the origin of that electricity based on the type of source (thermal, nuclear, renewable hydroelectric, or wind, among others), fuel consumption, water use, and the condition in which water is returned to the environment, etc. Some data are well known, but others—such as consumption data for a specific fleet of computers—are more difficult to obtain.
This project aims to develop a system that, by aggregating data from different sources and potentially using artificial intelligence techniques, can extrapolate this missing data and include it in the report. Language models (LLMs, SLMs) will also be used, complemented by RAG (Retrieval-Augmented Generation) techniques, to comply with the aforementioned reporting standards while minimizing the amount of human labor required.
The solution may take the form of a set of scripts, a PowerBI-style report visualization platform, or a web application, depending on the progress made in the project.
https://www.globalreporting.org/
- Complexity:9/10
- Technical specifications:LLM, SLM, RAG, AI
- CATEGORY 2024-2025 , Burgos Univ.
- TAGS C# , Python , Pytorch , React





