Foret

Inspiration

Social and environmental sustainability is become an increasingly important alpha for corporate and consumer investors alike. For corporate investors, a studying their portfolio's ESG ratings provides valuable insight into its ability to mitigate risk, capitalize on sustainability incentives, and sustain growth. For consumer investors, understanding a company's ESG rating allows them to buy stocks that align with their personal and social values. Thus, ESG data is an invaluable investment metric--companies already self-report long (70+ page!) ESG briefs, and data aggregators like Morningstar already sell quantitative ESG metrics to corporate investors.

However, as consumer investors, we've found that these 70-page technical ESG reports are difficult to understand, hard to read, and tend to be positively biased towards a company. Furthermore, quantitative ESG statistics are opaque and render little information. Thus, we wanted to create an application to not only provide short, qualitative ESG descriptions about a certain company, but also crosscheck the company's own report against third-party news articles and analysts.

What it does

Given an input company, Foret scrapes its ESG data from two sources: the company's own ESG report, and recent third-party news articles about the company. After getting this first-party and third-party data, Foret uses OCR and GPT technology to parse the data into short, bulleted summaries. Finally, Foret runs sentiment analysis on the third-party ESG news articles to create a "sustainability audit score" for the company.

We provide the user with side-by-side concise summaries of the company's ESG report and third-party ESG news coverage. This allows them to gain insight into the ESG goals of the company as well as how well they are achieving it.

We envision this tool to be embedded in consumer investing services like Robinhood to provide concise, fact-checked, qualitative sustainability summaries to aid new investors.

How we built it

To summarize a company's ESG report, we used Langchain to process their report PDF and summarize it into short plain text. Because ESG reports may be long, we used nested summarization to first summarize each page of the PDF, before pulling the key information from each page's summary.

To summarize the news coverage of a company's ESG metrics, we used the Aylien API to fetch news articles containing the company name and ESG keywords. We then summarized each article, generating a corresponding sentiment score. Afterwards, we used OpenAI's GPT API to summarize key points from the news article.

We used a Flask and React frontend to take user company inputs, then present a side-by-side view of the company and news ESG summaries.

Challenges we ran into

Merging our API with our react.js frontend--will still have to be finalized after the hackathon.

Scraping PDFs, running generative AI to summarize the models, and processing the sentiment of news articles takes a long time. After generating the summaries once, we cache them into a database for quick retrieval next time.

Selecting ESG keywords is difficult--ESG accounts for environmental, social, and intersectional sustainability metrics, so searching for the news articles to pull from is difficult.

Accomplishments that we're proud of

We are proud of the long data-processing pipeline we've built. From news articles and raw PDFs to concise, easy-to-read summaries, we are glad that our tool can fulfill our original metrics.

What we learned

We learned about the applications and implementations of generative AI models. Although we've interacted with ChatGPT for everyday questions, we now see its ability to impact financial and sustainability sectors.

We also learned how to maintain a large tech stack, and how to start with large unstructured web data like a PDF and process it down into a small summary.

What's next for Foret

Our next goal is to create an interactive website that allows consumer investors to easily harness the power of generative AI for their own ESG analysis. Therefore, they can quickly obtain information about a company's social impact. Our API could also be incorporated into other financial apps such as Robinhood.