The idea
PRISCA explores whether financial-news sentiment adds useful information to a next-day SPY opening-price forecast. Developed with a six-person AI Studio team, the project combines a price-data pipeline, NLP features, regression models, and a web interface.
The approach
Historical price features and news sentiment from VADER and FinBERT feed tree-based regression models. Model comparison and SHAP analysis help the team understand which inputs contribute to the forecast.
My contribution
- Contributed machine learning development, model training, and optimization.
- Helped coordinate the team as my role evolved from project management to technical leadership.
- Worked on feature preparation and evaluation alongside teammates responsible for sentiment analysis, data processing, and the application.
What I learned
The project reinforced the value of strong baselines and interpretable evaluation. Prior-price features were more influential than sentiment in the documented experiments.