This was a class project experimenting with using graph convolutional networks to predict the stationary distribution of queue networks. You can run through the code yourself here and read a write-up of the project here.

This was a project to track information flow across networks of LLM agents. Agents were given initial facts at the beginning of a "day", when would have conversations with connected agents in the graph. Agents were questioned on which information they knew after each round of conversation. You can read a write up of this project here. 

This study investigates the integration of Retrieval-Augmented Generation (RAG) models and ensemble methods to enhance the performance of Large Language Models (LLMs) in probabilistic forecasting tasks. Utilizing a novel dataset derived from forecasting platforms, we evaluate different LLMs enhanced by diverse retrieval strategies, including static and real-time data sources. Our results demonstrate that ensembling multiple RAG models significantly improves prediction accuracy and reliability. Future work will explore real-time data integration and advanced ensemble techniques to further refine the performance and applicability of RAG systems in dynamic forecasting environments. You can read a write up of this project here.