[absinha]

We are not using LLMs to replace the existing set of proven Machine Learning (ML) methods for extracting financial insights. Instead, LLMs play a key role in our system. They help reduce data overload and tailor the responses to the specific information an investor needs.

Consider a Bloomberg terminal - professional investors often spend days setting up their dashboards to see the right insights. With our system, investors can simply send natural language queries, and the LLMs act as a "glue layer" between the investor and the sophisticated quantitative models running behind the scenes.

Another major use case is screening. Traditional screening tools can be cumbersome, with hundreds of buttons and drop-downs. LLMs allow us to make the screening process more natural by generating the right queries from simple user questions. This is a more powerful approach.

In the past, sentiment analysis was challenging, requiring complex Named Entity Recognition, classification, and summarization models. Even basic 7 billion parameter LLMs now outperform hand-built classifiers for tasks like news sentiment analysis and SEC filings summarization.

In summary, our system is not just based on LLMs. It is a comprehensive solution that uses LLMs as a key component, alongside other essential parts like data ingestion pipelines and sophisticated quantitative models. The LLMs serve as an intuitive interface, allowing investors to access the power of the underlying financial analytics and lower the barrier to entries for all investors alike.