Filings Forecaster

Inspiration

Our fascination with Machine Learning (ML) and its transformative impact in data science led us to explore text generation through Natural Language Processing (NLP). The challenge of creating a predictive model using NLP techniques not only piqued our curiosity but also unraveled a myriad of practical applications in the financial sector.

What it does

Our project, the Filings Forecaster, adeptly predicts the footnotes of forthcoming 10-K annual filing reports by leveraging historical reports. This foresight can equip stakeholders with a preliminary understanding of a company's financial narrative.

How we built it

We employed a blend of cutting-edge technologies and methodologies to bring our idea to fruition. Utilizing the robustness of Langchain, the advanced text generation capability of GPT-4, and the versatility of Python, we trained our machine learning model on a dataset comprising past annual reports. Our backend was structured using Flask, which facilitated seamless interaction with the front-end designed using HTML, CSS, and a sprinkle of JavaScript for interactivity.

Challenges we ran into

Our initial hurdle was the model's inability to generate comprehensive predictions for future reports due to the lack of context, which resulted in limited information being forecasted. This challenge nudged us to further fine-tune our model by incorporating the available quarterly reports for 2023, thereby broadening its understanding and enhancing the prediction accuracy.

Among the hurdles, we navigated through issues encountered with Pinecone, a vector database essential for retrieving similar content, and faced rate limit errors with OpenAI during our model training phase. These challenges momentarily slowed our pace but propelled us to devise alternative strategies to ensure seamless data handling and model training, ultimately enriching our learning experience and project robustness.

Accomplishments that we're proud of

We take immense pride in not only achieving our primary objective but also in the learning trajectory that this project offered. We delved deeper into the nuances of NLP, honed our skills in Python, and explored the synergies between various ML frameworks and tools.

What's next for Filings Forecaster

With a successful prototype in place, we are enthusiastic about expanding our model's horizon. We aim to integrate more diverse datasets, such as responses from mental health questionnaires among middle school children, to uncover insightful trends. This expansion will allow us to explore how our model can provide value across different domains, marking a step towards a more data-informed and insightful future.