Earthlytics

Inspiration

Sustainability reports, though lengthy and complex, are essential for understanding the annual $33.7 billion corporate spending on sustainability (IDC). In an era where AI development requires increasingly high power consumption, tech companies are likely to boost sustainability efforts and focus on climate solutions to offset their environmental impact, making sustainability documents ever more important. While digging into the reports, we discovered a major obstacle: there is no effective way to glean important metrics and perform analysis. Sustainability data are often buried in lengthy official reports, presented in varied and complex formats. Extracting actionable insights from these documents is like finding a needle in a haystack. There is a gap in the market for a tool that could gather, simplify, and clarify these sustainability reports to make the data accessible and useful to anyone from investors to consumers. That's why we built Earthlytics: a platform designed to transform dense, inaccessible sustainability documents into clear, actionable insights with the help of Claude.

What it does

Earthlytics is a platform allowing users to search, visualize, and analyze sustainability data from corporations. We parsed globally accepted sustainability metrics such as ones from Global Reporting Initiative (GRI) & Sustainability Accounting Standards Board (SASB) for each company (water consumed, energy used, etc) from their sustainability documents for visualization and analysis. With Search, users can look up sustainability data of any top Fortune 500 tech company, get visualizations, and ask for specific insights from Claude's Sonnet. With Upload, users upload sustainability data in fields from any company of their choice on our platform, and also get insights. With Generation, the user fills in a few metrics and Claude will provide an outline.

How we built it

We scraped publicly available ESG reports from 100+ Fortune 500 tech companies. We used Anthropic's Claude on AWS Bedrock to parse pdfs for metrics among the GRI and SASB regulations into a JSON format and visualized them as radar and line plots on our platform. Each report is also vector embedded using Bedrock and stored in AWS Bedrock Knowledge Base, where they can be searched on our search engine. Users can use our custom fine-tuned Claude chatbot in real time to ask questions of each ESG report as well as ask for ways to improve.

Challenges we ran into

Parsing the sustainability documents and retrieving metrics using Claude in the correct format. Different ESG documents track different metrics and we initially had trouble deciding which ones to choose. We decided to benchmark the most concrete metrics outlined in globally accepted regulations for technology companies (GGE, Total Energy Consumed, Total Water Withdrawn etc), for better comparison. At times, we felt the urge to feature creep and implement additional sections to our project that were out of scope for our project.

What's next for Earthlytics:

We're committed to advancing Earthlytics further. As AI adoption accelerates, we anticipate a growing need for businesses to manage their increasing power consumption effectively. Simultaneously, we expect regulators to seek accessible, user-friendly methods to benchmark environmental impacts. Next, we look to beta test our product with ~five small businesses in the Berkeley area, which will allow us to get feedback and iterate.