First place AIM Hackathon: SAM - The Sustainability Advanced Model

 Authors: Emile Johnston, Muhammed Cakin, Terezia Olsiakova, Francesca Ricter

The hackathon, known for its focus on high-impact AI innovations, saw SAM rise to the top for its transformative approach to ESG management. This prototype – despite being built in only one day – was able to demonstrate how companies can better evaluate their compliance with sustainability goals, especially with regards to the European Green Deal.

The Challenge

Companies today face growing pressure to meet stringent ESG standards, particularly under new European Green Deal regulations aimed at achieving climate neutrality by 2050. However, many enterprises struggle with fragmented data, manual processes, and biased reporting systems, making it difficult to proactively manage their ESG obligations. This leads to delayed decision-making and increased risks of non-compliance. Andrei and Orion saw this gap as an opportunity to leverage AI for a smarter, more efficient way to handle ESG reporting—and SAM was born.

SAM is more than just an ESG reporting tool—it’s an AI-powered partner for businesses seeking to manage their sustainability efforts proactively. Built using LangChain, FAISS, and OpenAI’s GPT models, SAM gathers and analyzes real-time data from multiple sources, including sustainability reports and internet retrieval. It integrates web searches and uses vector databases to generate detailed ESG compliance results for selected companies. The platform offers a comprehensive ESG compliance score and actionable insights, helping companies optimize their sustainability strategies and stay competitive.

SAM’s system processes sustainability-related data using retrieval-augmented generation (RAG) and natural language processing (NLP), turning complex information into an easily digestible format. The platform follows a detailed workflow to achieve this:

1. ESG Reports are processed using PyPDF2 to extract relevant text.

2. The extracted text is then split and structured using LangChain, making it easier to analyze.

3. The structured data is transformed into embeddings using OpenAI's models, and then stored in a vector store powered by FAISS.

4. User Inputs, combined with EU Green Deal Criteria, initiate a prompt using LangChain, which is further enriched with internet searches (via DuckDuckGo) to add context and detail.

5. Finally, the LLM Query is executed using OpenAI’s GPT models to generate a compliance score and provide AI-driven insights.

This systematic approach allows SAM to perform both report-based and internet-based analyses, creating a complete view of a company's sustainability performance. Automated PDF processing downloads and extracts text from company reports, building searchable vector databases that make compliance assessment streamlined and precise. SAM’s interactive dashboards, built using Streamlit, provide intuitive, human-centric analyses tailored to enabling users to rapidly understand the state of sustainability in the companies of interest.