Inspiration

We thought prior work using LLMs for market sentiment analysis like FinBERT was interesting and we wanted to see if scaffolding to mimic real world social dynamics could improve market sentiment prediction.

What it does

We simulate Texan's reactions to news headlines about fifteen Texas relevant stock tickers (Tesla, Exxon Mobil, Dell etc)

How we built it

We came up with the idea after a couple hours of back and forth and then used Claude Code with the OMC scaffolding framework. Next we reviewed its design decisions and made small changes. Then we connected it to AWS and let the data pipeline run, and reviewed the results locally.

Challenges we ran into

We had to have Claude troubleshoot the code when we moved it to AWS because it needed to consider the rate at which we would be hitting public news / stock price APIs.'

Accomplishments that we're proud of

Using Claude Code.

What we learned

LLMs still struggle with novel idea generation but excel at implementation.

What's next for Sentiment Graph

We will publish the results openly on Arxiv.

We don't have a video available right now.

Built With

Share this project:

Updates